ref: bc2eebdf46a7de8b2927a0144dfd65bead6f1f15
dir: /doc/draft-ietf-codec-opus.xml/
<?xml version="1.0" encoding="utf-8"?> <!DOCTYPE rfc SYSTEM 'rfc2629.dtd'> <?rfc toc="yes" symrefs="yes" ?> <rfc ipr="trust200902" category="std" docName="draft-ietf-codec-opus-08"> <front> <title abbrev="Interactive Audio Codec">Definition of the Opus Audio Codec</title> <author initials="JM" surname="Valin" fullname="Jean-Marc Valin"> <organization>Mozilla</organization> <address> <postal> <street>650 Castro Street</street> <city>Mountain View</city> <region>CA</region> <code>94041</code> <country>USA</country> </postal> <phone>+1 650 903-0800</phone> <email>[email protected]</email> </address> </author> <author initials="K." surname="Vos" fullname="Koen Vos"> <organization>Skype Technologies S.A.</organization> <address> <postal> <street>Stadsgarden 6</street> <city>Stockholm</city> <region></region> <code>11645</code> <country>SE</country> </postal> <phone>+46 855 921 989</phone> <email>[email protected]</email> </address> </author> <author initials="T.B." surname="Terriberry" fullname="Timothy B. Terriberry"> <organization>Mozilla Corporation</organization> <address> <postal> <street>650 Castro Street</street> <city>Mountain View</city> <region>CA</region> <code>94041</code> <country>USA</country> </postal> <phone>+1 650 903-0800</phone> <email>[email protected]</email> </address> </author> <date day="16" month="August" year="2011" /> <area>General</area> <workgroup></workgroup> <abstract> <t> This document defines the Opus codec, designed for interactive speech and audio transmission over the Internet. </t> </abstract> </front> <middle> <section anchor="introduction" title="Introduction"> <t> The Opus codec is a real-time interactive audio codec designed to meet the requirements described in <xref target="requirements"></xref>. It is composed of a linear prediction (LP)-based layer and a Modified Discrete Cosine Transform (MDCT)-based layer. The main idea behind using two layers is that in speech, linear prediction techniques (such as CELP) code low frequencies more efficiently than transform (e.g., MDCT) domain techniques, while the situation is reversed for music and higher speech frequencies. Thus a codec with both layers available can operate over a wider range than either one alone and, by combining them, achieve better quality than either one individually. </t> <t> The primary normative part of this specification is provided by the source code in <xref target="ref-implementation"></xref>. In general, only the decoder portion of this software is normative, though a significant amount of code is shared by both the encoder and decoder. <!--TODO: Forward reference conformance test--> The decoder contains significant amounts of integer and fixed-point arithmetic which must be performed exactly, including all rounding considerations, so any useful specification must make extensive use of domain-specific symbolic language to adequately define these operations. Additionally, any conflict between the symbolic representation and the included reference implementation must be resolved. For the practical reasons of compatibility and testability it would be advantageous to give the reference implementation priority in any disagreement. The C language is also one of the most widely understood human-readable symbolic representations for machine behavior. For these reasons this RFC uses the reference implementation as the sole symbolic representation of the codec. </t> <!--TODO: C is not unambiguous; many parts are implementation-defined--> <t>While the symbolic representation is unambiguous and complete it is not always the easiest way to understand the codec's operation. For this reason this document also describes significant parts of the codec in English and takes the opportunity to explain the rationale behind many of the more surprising elements of the design. These descriptions are intended to be accurate and informative, but the limitations of common English sometimes result in ambiguity, so it is expected that the reader will always read them alongside the symbolic representation. Numerous references to the implementation are provided for this purpose. The descriptions sometimes differ from the reference in ordering or through mathematical simplification wherever such deviation makes an explanation easier to understand. For example, the right shift and left shift operations in the reference implementation are often described using division and multiplication in the text. In general, the text is focused on the "what" and "why" while the symbolic representation most clearly provides the "how". </t> <section anchor="notation" title="Notation and Conventions"> <t> The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119. </t> <t> Even when using floating-point, various operations in the codec require bit-exact fixed-point behavior. The notation "Q<n>", where n is an integer, denotes the number of binary digits to the right of the decimal point in a fixed-point number. For example, a signed Q14 value in a 16-bit word can represent values from -2.0 to 1.99993896484375, inclusive. This notation is for informational purposes only. Arithmetic, when described, always operates on the underlying integer. E.g., the text will explicitly indicate any shifts required after a multiplication. </t> <t> Expressions, where included in the text, follow C operator rules and precedence, with the exception that syntax like "2**n" is used to indicate 2 raised to the power n. The text also makes use of the following functions: </t> <section anchor="min" title="min(x,y)"> <t> The smallest of two values x and y. </t> </section> <section anchor="max" title="max(x,y)"> <t> The largest of two values x and y. </t> </section> <section anchor="clamp" title="clamp(lo,x,hi)"> <figure align="center"> <artwork align="center"><![CDATA[ clamp(lo,x,hi) = max(lo,min(x,hi)) ]]></artwork> </figure> <t> With this definition, if lo>hi, the lower bound is the one that is enforced. </t> </section> <section anchor="sign" title="sign(x)"> <t> The sign of x, i.e., <figure align="center"> <artwork align="center"><![CDATA[ ( -1, x < 0 , sign(x) = < 0, x == 0 , ( 1, x > 0 . ]]></artwork> </figure> </t> </section> <section anchor="log2" title="log2(f)"> <t> The base-two logarithm of f. </t> </section> <section anchor="ilog" title="ilog(n)"> <t> The minimum number of bits required to store a positive integer n in two's complement notation, or 0 for a non-positive integer n. <figure align="center"> <artwork align="center"><![CDATA[ ( 0, n <= 0, ilog(n) = < ( floor(log2(n))+1, n > 0 ]]></artwork> </figure> Examples: <list style="symbols"> <t>ilog(-1) = 0</t> <t>ilog(0) = 0</t> <t>ilog(1) = 1</t> <t>ilog(2) = 2</t> <t>ilog(3) = 2</t> <t>ilog(4) = 3</t> <t>ilog(7) = 3</t> </list> </t> </section> </section> </section> <section anchor="overview" title="Opus Codec Overview"> <t> The Opus codec scales from 6 kb/s narrowband mono speech to 510 kb/s fullband stereo music, with algorithmic delays ranging from 5 ms to 65.2 ms. At any given time, either the LP layer, the MDCT layer, or both, may be active. It can seamlessly switch between all of its various operating modes, giving it a great deal of flexibility to adapt to varying content and network conditions without renegotiating the current session. Internally, the codec always operates at a 48 kHz sampling rate, though it allows input and output of various bandwidths, defined as follows: </t> <texttable> <ttcol>Abbreviation</ttcol> <ttcol align="right">Audio Bandwidth</ttcol> <ttcol align="right">Sampling Rate (Effective)</ttcol> <c>NB (narrowband)</c> <c>4 kHz</c> <c>8 kHz</c> <c>MB (medium-band)</c> <c>6 kHz</c> <c>12 kHz</c> <c>WB (wideband)</c> <c>8 kHz</c> <c>16 kHz</c> <c>SWB (super-wideband)</c> <c>12 kHz</c> <c>24 kHz</c> <c>FB (fullband)</c> <c>20 kHz</c> <c>48 kHz</c> </texttable> <t> These can be chosen independently on the encoder and decoder side, e.g., a fullband signal can be decoded as wideband, or vice versa. This approach ensures a sender and receiver can always interoperate, regardless of the capabilities of their actual audio hardware. </t> <t> Opus defines super-wideband (SWB) mode to have an effective sampling rate of 24 kHz, unlike some other audio coding standards that use 32 kHz. This was chosen for a number of reasons. The band layout in the MDCT layer naturally allows skipping coefficients for frequencies over 12 kHz, but does not allow cleanly dropping frequencies over 16 kHz. The choice of 24 kHz also makes resampling in the MDCT layer easier, as 24 evenly divides 48, and when 24 kHz is sufficient, it can save computation in other processing, such as Acoustic Echo Cancellation (AEC). Experimental changes to the band layout to allow a 16 kHz cutoff showed potential quality degredations, and at typical bitrates the number of bits saved by using such a cutoff instead of coding in fullband (FB) mode is very small. Therefore, if an application wishes to process a signal sampled at 32 kHz, it should just use FB mode. </t> <t> The LP layer is based on the <eref target='http://developer.skype.com/silk'>SILK</eref> codec <xref target="SILK"></xref>. It supports NB, MB, or WB audio and frame sizes from 10 ms to 60 ms, and requires an additional 5.2 ms look-ahead for noise shaping estimation (5 ms) and internal resampling (0.2 ms). Like Vorbis and many other modern codecs, SILK is inherently designed for variable-bitrate (VBR) coding, though an encoder can with sufficient effort produce constant-bitrate (CBR) or near-CBR streams. </t> <t> The MDCT layer is based on the <eref target='http://www.celt-codec.org/'>CELT</eref> codec <xref target="CELT"></xref>. It supports sampling NB, WB, SWB, or FB audio and frame sizes from 2.5 ms to 20 ms, and requires an additional 2.5 ms look-ahead due to the overlapping MDCT windows. The CELT codec is inherently designed for CBR coding, but unlike many CBR codecs it is not limited to a set of predetermined rates. It internally allocates bits to exactly fill any given target budget, and an encoder can produce a VBR stream by varying the target on a per-frame basis. The MDCT layer is not used for speech when the audio bandwidth is WB or less, as it is not useful there. On the other hand, non-speech signals are not always adequately coded using linear prediction, so for music only the MDCT layer should be used. </t> <t> A hybrid mode allows the use of both layers simultaneously with a frame size of 10 or 20 ms and a SWB or FB audio bandwidth. Each frame is split into a low frequency signal and a high frequency signal, with a cutoff of 8 kHz. The LP layer then codes the low frequency signal, followed by the MDCT layer coding the high frequency signal. In the MDCT layer, all bands below 8 kHz are discarded, so there is no coding redundancy between the two layers. </t> <t> At the decoder, the two decoder outputs are simply added together. To compensate for the different look-aheads required by each layer, the CELT encoder input is delayed by an additional 2.7 ms. This ensures that low frequencies and high frequencies arrive at the same time. This extra delay MAY be reduced by an encoder by using less look-ahead for noise shaping or using a simpler resampler in the LP layer, but this will reduce quality. However, the base 2.5 ms look-ahead in the CELT layer cannot be reduced in the encoder because it is needed for the MDCT overlap, whose size is fixed by the decoder. </t> <t> Both layers use the same entropy coder, avoiding any waste from "padding bits" between them. The hybrid approach makes it easy to support both CBR and VBR coding. Although the LP layer is VBR, the bit allocation of the MDCT layer can produce a final stream that is CBR by using all the bits left unused by the LP layer. </t> </section> <section anchor="modes" title="Codec Modes"> <t> As described, the two layers can be combined in three possible operating modes: <list style="numbers"> <t>A LP-only mode for use in low bitrate connections with an audio bandwidth of WB or less,</t> <t>A hybrid (LP+MDCT) mode for SWB or FB speech at medium bitrates, and</t> <t>An MDCT-only mode for very low delay speech transmission as well as music transmission.</t> </list> </t> <t> A single packet may contain multiple audio frames. However, they must share a common set of parameters, including the operating mode, audio bandwidth, frame size, and channel count. This section describes the possible combinations of these parameters and the internal framing used to pack multiple frames into a single packet. This framing is not self-delimiting. Instead, it assumes that a higher layer (such as UDP or RTP or Ogg or Matroska) will communicate the length, in bytes, of the packet, and it uses this information to reduce the framing overhead in the packet itself. A decoder implementation MUST support the framing described in this section. An alternative, self-delimiting variant of the framing is described in <xref target="self-delimiting-framing"/>. Support for that variant is OPTIONAL. </t> <section anchor="toc_byte" title="The TOC Byte"> <t> An Opus packet begins with a single-byte table-of-contents (TOC) header that signals which of the various modes and configurations a given packet uses. It is composed of a frame count code, "c", a stereo flag, "s", and a configuration number, "config", arranged as illustrated in <xref target="toc_byte_fig"/>. A description of each of these fields follows. </t> <figure anchor="toc_byte_fig" title="The TOC byte"> <artwork align="center"><![CDATA[ 0 0 1 2 3 4 5 6 7 +-+-+-+-+-+-+-+-+ | c |s| config | +-+-+-+-+-+-+-+-+ ]]></artwork> </figure> <t> The top five bits of the TOC byte, labeled "config", encode one of 32 possible configurations of operating mode, audio bandwidth, and frame size. <xref target="config_bits"/> lists the parameters for each configuration. </t> <texttable anchor="config_bits" title="TOC Byte Configuration Parameters"> <ttcol>Configuration Number(s)</ttcol> <ttcol>Mode</ttcol> <ttcol>Bandwidth</ttcol> <ttcol>Frame Size(s)</ttcol> <c>0...3</c> <c>LP-only</c> <c>NB</c> <c>10, 20, 40, 60 ms</c> <c>4...7</c> <c>LP-only</c> <c>MB</c> <c>10, 20, 40, 60 ms</c> <c>8...11</c> <c>LP-only</c> <c>WB</c> <c>10, 20, 40, 60 ms</c> <c>12...13</c> <c>Hybrid</c> <c>SWB</c> <c>10, 20 ms</c> <c>14...15</c> <c>Hybrid</c> <c>FB</c> <c>10, 20 ms</c> <c>16...19</c> <c>MDCT-only</c> <c>NB</c> <c>2.5, 5, 10, 20 ms</c> <c>20...23</c> <c>MDCT-only</c> <c>WB</c> <c>2.5, 5, 10, 20 ms</c> <c>24...27</c> <c>MDCT-only</c> <c>SWB</c> <c>2.5, 5, 10, 20 ms</c> <c>28...31</c> <c>MDCT-only</c> <c>FB</c> <c>2.5, 5, 10, 20 ms</c> </texttable> <t> One additional bit, labeled "s", is used to signal mono vs. stereo, with 0 indicating mono and 1 indicating stereo. </t> <t> The remaining two bits of the TOC byte, labeled "c", code the number of frames per packet (codes 0 to 3) as follows: <list style="symbols"> <t>0: 1 frame in the packet</t> <t>1: 2 frames in the packet, each with equal compressed size</t> <t>2: 2 frames in the packet, with different compressed sizes</t> <t>3: an arbitrary number of frames in the packet</t> </list> This draft refers to a packet as a code 0 packet, code 1 packet, etc., based on the value of "c". </t> <t> A well-formed Opus packet MUST contain at least one byte with the TOC information, though the frame(s) within a packet MAY be zero bytes long. </t> </section> <section title="Frame Packing"> <t> This section describes how frames are packed according to each possible value of "c" in the TOC byte. </t> <section anchor="frame-length-coding" title="Frame Length Coding"> <t> When a packet contains multiple VBR frames, the compressed length of one or more of these frames is indicated with a one or two byte sequence, with the meaning of the first byte as follows: <list style="symbols"> <t>0: No frame (discontinuous transmission (DTX) or lost packet)</t> <!--TODO: Would be nice to be clearer about the distinction between "frame size" (in samples or ms) and "the compressed size of the frame" (in bytes). "the compressed length of the frame" is maybe a little better, but not when we jump back and forth to talking about sizes.--> <t>1...251: Size of the frame in bytes</t> <t>252...255: A second byte is needed. The total size is (size[1]*4)+size[0]</t> </list> </t> <t> The maximum representable size is 255*4+255=1275 bytes. This limit MUST NOT be exceeded, even when no length field is used. For 20 ms frames, this represents a bitrate of 510 kb/s, which is approximately the highest useful rate for lossily compressed fullband stereo music. Beyond this point, lossless codecs are more appropriate. It is also roughly the maximum useful rate of the MDCT layer, as shortly thereafter quality no longer improves with additional bits due to limitations on the codebook sizes. </t> <t> No length is transmitted for the last frame in a VBR packet, or for any of the frames in a CBR packet, as it can be inferred from the total size of the packet and the size of all other data in the packet. However, the length of any individual frame MUST NOT exceed 1275 bytes, to allow for repacketization by gateways, conference bridges, or other software. </t> </section> <section title="One Frame in the Packet (Code 0)"> <t> For code 0 packets, the TOC byte is immediately followed by N-1 bytes of compressed data for a single frame (where N is the size of the packet), as illustrated in <xref target="code0_packet"/>. </t> <figure anchor="code0_packet" title="A Code 0 Packet" align="center"> <artwork align="center"><![CDATA[ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |0|0|s| config | | +-+-+-+-+-+-+-+-+ | | Compressed frame 1 (N-1 bytes)... : : | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> </section> <section title="Two Frames in the Packet, Each with Equal Compressed Size (Code 1)"> <t> For code 1 packets, the TOC byte is immediately followed by the (N-1)/2 bytes of compressed data for the first frame, followed by (N-1)/2 bytes of compressed data for the second frame, as illustrated in <xref target="code1_packet"/>. The number of payload bytes available for compressed data, N-1, MUST be even for all code 1 packets. </t> <figure anchor="code1_packet" title="A Code 1 Packet" align="center"> <artwork align="center"><![CDATA[ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|0|s| config | | +-+-+-+-+-+-+-+-+ : | Compressed frame 1 ((N-1)/2 bytes)... | : +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ : | Compressed frame 2 ((N-1)/2 bytes)... | : +-+-+-+-+-+-+-+-+ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> </section> <section title="Two Frames in the Packet, with Different Compressed Sizes (Code 2)"> <t> For code 2 packets, the TOC byte is followed by a one or two byte sequence indicating the length of the first frame (marked N1 in the figure below), followed by N1 bytes of compressed data for the first frame. The remaining N-N1-2 or N-N1-3 bytes are the compressed data for the second frame. This is illustrated in <xref target="code2_packet"/>. The length of the first frame, N1, MUST be no larger than the size of the payload remaining after decoding that length for all code 2 packets. </t> <figure anchor="code2_packet" title="A Code 2 Packet" align="center"> <artwork align="center"><![CDATA[ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |0|1|s| config | N1 (1-2 bytes): | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ : | Compressed frame 1 (N1 bytes)... | : +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | Compressed frame 2... : : | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> </section> <section title="An Arbitrary Number of Frames in the Packet (Code 3)"> <t> Code 3 packets may encode an arbitrary number of frames, as well as additional padding, called "Opus padding" to indicate that this padding is added at the Opus layer, rather than at the transport layer. For code 3 packets, the TOC byte is followed by a byte encoding the number of frames in the packet in bits 0 to 5 (marked "M" in the figure below), with bit 6 indicating whether or not Opus padding is inserted (marked "p" in the figure below), and bit 7 indicating VBR (marked "v" in the figure below). M MUST NOT be zero, and the audio duration contained within a packet MUST NOT exceed 120 ms. This limits the maximum frame count for any frame size to 48 (for 2.5 ms frames), with lower limits for longer frame sizes. <xref target="frame_count_byte"/> illustrates the layout of the frame count byte. </t> <figure anchor="frame_count_byte" title="The frame count byte"> <artwork align="center"><![CDATA[ 0 0 1 2 3 4 5 6 7 +-+-+-+-+-+-+-+-+ | M |p|v| +-+-+-+-+-+-+-+-+ ]]></artwork> </figure> <t> When Opus padding is used, the number of bytes of padding is encoded in the bytes following the frame count byte. Values from 0...254 indicate that 0...254 bytes of padding are included, in addition to the byte(s) used to indicate the size of the padding. If the value is 255, then the size of the additional padding is 254 bytes, plus the padding value encoded in the next byte. The additional padding bytes appear at the end of the packet, and SHOULD be set to zero by the encoder. The decoder MUST accept any value for the padding bytes, however. By using code 255 multiple times, it is possible to create a packet of any specific, desired size. Let P be the total amount of padding, including both the trailing padding bytes themselves and the header bytes used to indicate how many there are. Then P MUST be no more than N-2 for CBR packets, or N-M-1 for VBR packets. </t> <t> In the CBR case, the compressed length of each frame in bytes is equal to the number of remaining bytes in the packet after subtracting the (optional) padding, (N-2-P), divided by M. This number MUST be an integer multiple of M. The compressed data for all M frames then follows, each of size (N-2-P)/M bytes, as illustrated in <xref target="code3cbr_packet"/>. </t> <figure anchor="code3cbr_packet" title="A CBR Code 3 Packet" align="center"> <artwork align="center"><![CDATA[ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|1|s| config | M |p|0| Padding length (Optional) : +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : Compressed frame 1 ((N-2-P)/M bytes)... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : Compressed frame 2 ((N-2-P)/M bytes)... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : ... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : Compressed frame M ((N-2-P)/M bytes)... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ : Opus Padding (Optional)... | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> <t> In the VBR case, the (optional) padding length is followed by M-1 frame lengths (indicated by "N1" to "N[M-1]" in the figure below), each encoded in a one or two byte sequence as described above. The packet MUST contain enough data for the M-1 lengths after the (optional) padding, and the sum of these lengths MUST be no larger than the number of bytes remaining in the packet after decoding them. The compressed data for all M frames follows, each frame consisting of the indicated number of bytes, with the final frame consuming any remaining bytes before the final padding, as illustrated in <xref target="code3cbr_packet"/>. The number of header bytes (TOC byte, frame count byte, padding length bytes, and frame length bytes), plus the length of the first M-1 frames themselves, plus the length of the padding MUST be no larger than N, the total size of the packet. </t> <figure anchor="code3vbr_packet" title="A VBR Code 3 Packet" align="center"> <artwork align="center"><![CDATA[ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|1|s| config | M |p|1| Padding length (Optional) : +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ : N1 (1-2 bytes): N2 (1-2 bytes): ... : N[M-1] | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : Compressed frame 1 (N1 bytes)... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : Compressed frame 2 (N2 bytes)... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : ... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : Compressed frame M... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ : Opus Padding (Optional)... | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> </section> </section> <section anchor="examples" title="Examples"> <t> Simplest case, one NB mono 20 ms SILK frame: </t> <figure> <artwork><![CDATA[ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |0|0|0| 1 | compressed data... : +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> <t> Two FB mono 5 ms CELT frames of the same compressed size: </t> <figure> <artwork><![CDATA[ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|0|0| 29 | compressed data... : +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> <t> Two FB mono 20 ms hybrid frames of different compressed size: </t> <figure> <artwork><![CDATA[ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|1|0| 15 | 2 |0|1| N1 | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | compressed data... : +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> <t> Four FB stereo 20 ms CELT frames of the same compressed size: </t> <figure> <artwork><![CDATA[ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|1|1| 31 | 4 |0|0| compressed data... : +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> </section> <section title="Extending Opus"> <t> A receiver MUST NOT process packets which violate the rules above as normal Opus packets. They are reserved for future applications, such as in-band headers (containing metadata, etc.) or multichannel support. </t> </section> </section> <section title="Opus Decoder"> <t> The Opus decoder consists of two main blocks: the SILK decoder and the CELT decoder. The output of the Opus decode is the sum of the outputs from the SILK and CELT decoders with proper sample rate conversion and delay compensation as illustrated in the block diagram below. At any given time, one or both of the SILK and CELT decoders may be active. </t> <figure> <artwork> <![CDATA[ +-------+ +----------+ | SILK | | sample | +->|decoder|--->| rate |----+ bit- +-------+ | | | |conversion| v stream | Range |---+ +-------+ +----------+ /---\ audio ------->|decoder| | + |------> | |---+ +-------+ \---/ +-------+ | | CELT | ^ +-------------->|decoder|-------+ | | +-------+ ]]> </artwork> </figure> <section anchor="range-decoder" title="Range Decoder"> <t> Opus uses an entropy coder based on <xref target="range-coding"></xref>, which is itself a rediscovery of the FIFO arithmetic code introduced by <xref target="coding-thesis"></xref>. It is very similar to arithmetic encoding, except that encoding is done with digits in any base instead of with bits, so it is faster when using larger bases (i.e., an octet). All of the calculations in the range coder must use bit-exact integer arithmetic. </t> <t> Symbols may also be coded as "raw bits" packed directly into the bitstream, bypassing the range coder. These are packed backwards starting at the end of the frame, as illustrated in <xref target="rawbits-example"/>. This reduces complexity and makes the stream more resilient to bit errors, as corruption in the raw bits will not desynchronize the decoding process, unlike corruption in the input to the range decoder. Raw bits are only used in the CELT layer. </t> <figure anchor="rawbits-example" title="Illustrative example of packing range coder and raw bits data"> <artwork align="center"><![CDATA[ 0 1 2 3 7 6 5 4 3 2 1 0 7 6 5 4 3 2 1 0 7 6 5 4 3 2 1 0 7 6 5 4 3 2 1 0 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Range coder data (packed MSb to LSb) -> : + + : : + +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ : | <- Boundary occurs at an arbitrary bit position : +-+-+-+ + : <- Raw bits data (packed LSb to MSb) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> <t> Each symbol coded by the range coder is drawn from a finite alphabet and coded in a separate "context", which describes the size of the alphabet and the relative frequency of each symbol in that alphabet. </t> <t> Suppose there is a context with n symbols, identified with an index that ranges from 0 to n-1. The parameters needed to encode or decode a symbol in this context are represented by a three-tuple (fl[k], fh[k], ft), with 0 <= fl[k] < fh[k] <= ft <= 65535. The values of this tuple are derived from the probability model for the symbol, represented by traditional "frequency counts". Because Opus uses static contexts these are not updated as symbols are decoded. Let f[i] be the frequency of symbol i. Then the three-tuple corresponding to symbol k is given by </t> <figure align="center"> <artwork align="center"><![CDATA[ k-1 n-1 __ __ fl[k] = \ f[i], fh[k] = fl[k] + f[k], ft[k] = \ f[i] /_ /_ i=0 i=0 ]]></artwork> </figure> <t> The range decoder extracts the symbols and integers encoded using the range encoder in <xref target="range-encoder"/>. The range decoder maintains an internal state vector composed of the two-tuple (val,rng), representing the difference between the high end of the current range and the actual coded value, minus one, and the size of the current range, respectively. Both val and rng are 32-bit unsigned integer values. The decoder initializes rng to 128 and initializes val to 127 minus the top 7 bits of the first input octet. The remaining bit is saved for use in the renormalization procedure described in <xref target="range-decoder-renorm"/>, which the decoder invokes immediately after initialization to read additional bits and establish the invariant that rng > 2**23. </t> <section anchor="decoding-symbols" title="Decoding Symbols"> <t> Decoding a symbol is a two-step process. The first step determines a 16-bit unsigned value fs, which lies within the range of some symbol in the current context. The second step updates the range decoder state with the three-tuple (fl[k], fh[k], ft) corresponding to that symbol. </t> <t> The first step is implemented by ec_decode() (entdec.c), which computes <figure align="center"> <artwork align="center"><![CDATA[ fs = ft - min(val/(rng/ft)+1, ft) . ]]></artwork> </figure> The divisions here are exact integer division. </t> <t> The decoder then identifies the symbol in the current context corresponding to fs; i.e., the value of k whose three-tuple (fl[k], fh[k], ft) satisfies fl[k] <= fs < fh[k]. It uses this tuple to update val according to <figure align="center"> <artwork align="center"><![CDATA[ val = val - (rng/ft)*(ft-fh[k]) . ]]></artwork> </figure> If fl[k] is greater than zero, then the decoder updates rng using <figure align="center"> <artwork align="center"><![CDATA[ rng = (rng/ft)*(fh[k]-fl[k]) . ]]></artwork> </figure> Otherwise, it updates rng using <figure align="center"> <artwork align="center"><![CDATA[ rng = rng - (rng/ft)*(ft-fh[k]). ]]></artwork> </figure> </t> <t> Using a special case for the first symbol (rather than the last symbol, as is commonly done in other arithmetic coders) ensures that all the truncation error from the finite precision arithmetic accumulates in symbol 0. This makes the cost of coding a 0 slightly smaller, on average, than its estimated probability indicates and makes the cost of coding any other symbol slightly larger. When contexts are designed so that 0 is the most probable symbol, which is often the case, this strategy minimizes the inefficiency introduced by the finite precision. It also makes some of the special-case decoding routines in <xref target="decoding-alternate"/> particularly simple. </t> <t> After the updates, implemented by ec_dec_update() (entdec.c), the decoder normalizes the range using the procedure in the next section, and returns the index k. </t> <section anchor="range-decoder-renorm" title="Renormalization"> <t> To normalize the range, the decoder repeats the following process, implemented by ec_dec_normalize() (entdec.c), until rng > 2**23. If rng is already greater than 2**23, the entire process is skipped. First, it sets rng to (rng<<8). Then it reads the next octet of the payload and combines it with the left-over bit buffered from the previous octet to form the 8-bit value sym. It takes the left-over bit as the high bit (bit 7) of sym, and the top 7 bits of the octet it just read as the other 7 bits of sym. The remaining bit in the octet just read is buffered for use in the next iteration. If no more input octets remain, it uses zero bits instead. Then, it sets <figure align="center"> <artwork align="center"><![CDATA[ val = ((val<<8) + (255-sym)) & 0x7FFFFFFF . ]]></artwork> </figure> </t> <t> It is normal and expected that the range decoder will read several bytes into the raw bits data (if any) at the end of the packet by the time the frame is completely decoded, as illustrated in <xref target="finalize-example"/>. This same data MUST also be returned as raw bits when requested. The encoder is expected to terminate the stream in such a way that the decoder will decode the intended values regardless of the data contained in the raw bits. <xref target="encoder-finalizing"/> describes a procedure for doing this. If the range decoder consumes all of the bytes belonging to the current frame, it MUST continue to use zero when any further input bytes are required, even if there is additional data in the current packet from padding or other frames. </t> <figure anchor="finalize-example" title="Illustrative example of raw bits overlapping range coder data"> <artwork align="center"><![CDATA[ n n+1 n+2 n+3 7 6 5 4 3 2 1 0 7 6 5 4 3 2 1 0 7 6 5 4 3 2 1 0 7 6 5 4 3 2 1 0 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ : | <----------- Overlap region ------------> | : +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ^ ^ | End of data buffered by the range coder | ...-----------------------------------------------+ | | End of data consumed by raw bits +-------------------------------------------------------... ]]></artwork> </figure> </section> </section> <section anchor="decoding-alternate" title="Alternate Decoding Methods"> <t> The reference implementation uses three additional decoding methods that are exactly equivalent to the above, but make assumptions and simplifications that allow for a more efficient implementation. </t> <section title="ec_decode_bin()"> <t> The first is ec_decode_bin() (entdec.c), defined using the parameter ftb instead of ft. It is mathematically equivalent to calling ec_decode() with ft = (1<<ftb), but avoids one of the divisions. </t> </section> <section title="ec_dec_bit_logp()"> <t> The next is ec_dec_bit_logp() (entdec.c), which decodes a single binary symbol, replacing both the ec_decode() and ec_dec_update() steps. The context is described by a single parameter, logp, which is the absolute value of the base-2 logarithm of the probability of a "1". It is mathematically equivalent to calling ec_decode() with ft = (1<<logp), followed by ec_dec_update() with the 3-tuple (fl[k] = 0, fh[k] = (1<<logp)-1, ft = (1<<logp)) if the returned value of fs is less than (1<<logp)-1 (a "0" was decoded), and with (fl[k] = (1<<logp)-1, fh[k] = ft = (1<<logp)) otherwise (a "1" was decoded). The implementation requires no multiplications or divisions. </t> </section> <section title="ec_dec_icdf()"> <t> The last is ec_dec_icdf() (entdec.c), which decodes a single symbol with a table-based context of up to 8 bits, also replacing both the ec_decode() and ec_dec_update() steps, as well as the search for the decoded symbol in between. The context is described by two parameters, an icdf ("inverse" cumulative distribution function) table and ftb. As with ec_decode_bin(), (1<<ftb) is equivalent to ft. idcf[k], on the other hand, stores (1<<ftb)-fh[k], which is equal to (1<<ftb)-fl[k+1]. fl[0] is assumed to be 0, and the table is terminated by a value of 0 (where fh[k] == ft). </t> <t> The function is mathematically equivalent to calling ec_decode() with ft = (1<<ftb), using the returned value fs to search the table for the first entry where fs < (1<<ftb)-icdf[k], and calling ec_dec_update() with fl[k] = (1<<ftb)-icdf[k-1] (or 0 if k == 0), fh[k] = (1<<ftb)-idcf[k], and ft = (1<<ftb). Combining the search with the update allows the division to be replaced by a series of multiplications (which are usually much cheaper), and using an inverse CDF allows the use of an ftb as large as 8 in an 8-bit table without any special cases. This is the primary interface with the range decoder in the SILK layer, though it is used in a few places in the CELT layer as well. </t> <t> Although icdf[k] is more convenient for the code, the frequency counts, f[k], are a more natural representation of the probability distribution function (PDF) for a given symbol. Therefore this draft lists the latter, not the former, when describing the context in which a symbol is coded as a list, e.g., {4, 4, 4, 4}/16 for a uniform context with four possible values and ft=16. The value of ft after the slash is always the sum of the entries in the PDF, but is included for convenience. Contexts with identical probabilities, f[k]/ft, but different values of ft (or equivalently, ftb) are not the same, and cannot, in general, be used in place of one another. An icdf table is also not capable of representing a PDF where the first symbol has 0 probability. In such contexts, ec_dec_icdf() can decode the symbol by using a table that drops the entries for any initial zero-probability values and adding the constant offset of the first value with a non-zero probability to its return value. </t> </section> </section> <section anchor="decoding-bits" title="Decoding Raw Bits"> <t> The raw bits used by the CELT layer are packed at the end of the packet, with the least significant bit of the first value packed in the least significant bit of the last byte, filling up to the most significant bit in the last byte, continuing on to the least significant bit of the penultimate byte, and so on. The reference implementation reads them using ec_dec_bits() (entdec.c). Because the range decoder must read several bytes ahead in the stream, as described in <xref target="range-decoder-renorm"/>, the input consumed by the raw bits MAY overlap with the input consumed by the range coder, and a decoder MUST allow this. The format should render it impossible to attempt to read more raw bits than there are actual bits in the frame, though a decoder MAY wish to check for this and report an error. </t> </section> <section anchor="decoding-ints" title="Decoding Uniformly Distributed Integers"> <t> The ec_dec_uint() (entdec.c) function decodes one of ft equiprobable values in the range 0 to ft-1, inclusive, each with a frequency of 1, where ft may be as large as 2**32-1. Because ec_decode() is limited to a total frequency of 2**16-1, this is split up into a range coded symbol representing up to 8 of the high bits of the value, and, if necessary, raw bits representing the remaining bits. The limit of 8 bits in the range coded symbol is a trade-off between implementation complexity, modeling error (since the symbols no longer truly have equal coding cost), and rounding error introduced by the range coder itself (which gets larger as more bits are included). Using raw bits reduces the maximum number of divisions required in the worst case, but means that it may be possible to decode a value outside the range 0 to ft-1, inclusive. </t> <t> ec_dec_uint() takes a single, positive parameter, ft, which is not necessarily a power of two, and returns an integer, t, whose value lies between 0 and ft-1, inclusive. Let ftb = ilog(ft-1), i.e., the number of bits required to store ft-1 in two's complement notation. If ftb is 8 or less, then t is decoded with t = ec_decode(ft), and the range coder state is updated using the three-tuple (t,t+1,ft). </t> <t> If ftb is greater than 8, then the top 8 bits of t are decoded using t = ec_decode((ft-1>>ftb-8)+1), the decoder state is updated using the three-tuple (t,t+1,(ft-1>>ftb-8)+1), and the remaining bits are decoded as raw bits, setting t = t<<ftb-8|ec_dec_bits(ftb-8). If, at this point, t >= ft, then the current frame is corrupt. In that case, the decoder should assume there has been an error in the coding, decoding, or transmission and SHOULD take measures to conceal the error and/or report to the application that a problem has occurred. </t> </section> <section anchor="decoder-tell" title="Current Bit Usage"> <t> The bit allocation routines in the CELT decoder need a conservative upper bound on the number of bits that have been used from the current frame thus far, including both range coder bits and raw bits. This drives allocation decisions that must match those made in the encoder. The upper bound is computed in the reference implementation to whole-bit precision by the function ec_tell() (entcode.h) and to fractional 1/8th bit precision by the function ec_tell_frac() (entcode.c). Like all operations in the range coder, it must be implemented in a bit-exact manner, and must produce exactly the same value returned by the same functions in the encoder after encoding the same symbols. </t> <t> ec_tell() is guaranteed to return ceil(ec_tell_frac()/8.0). In various places the codec will check to ensure there is enough room to contain a symbol before attempting to decode it. In practice, although the number of bits used so far is an upper bound, decoding a symbol whose probability model suggests it has a worst-case cost of p 1/8th bits may actually advance the return value of ec_tell_frac() by p-1, p, or p+1 1/8th bits, due to approximation error in that upper bound, truncation error in the range coder, and for large values of ft, modeling error in ec_dec_uint(). </t> <t> However, this error is bounded, and periodic calls to ec_tell() or ec_tell_frac() at precisely defined points in the decoding process prevent it from accumulating. For a range coder symbol that requires a whole number of bits (i.e., for which ft/(fh[k]-fl[k]) is a power of two), where there are at least p 1/8th bits available, decoding the symbol will never cause ec_tell() or ec_tell_frac() to exceed the size of the frame ("bust the budget"). In this case the return value of ec_tell_frac() will only advance by more than p 1/8th bits if there was an additional, fractional number of bits remaining, and it will never advance beyond the next whole-bit boundary, which is safe, since frames always contain a whole number of bits. However, when p is not a whole number of bits, an extra 1/8th bit is required to ensure that decoding the symbol will not bust the budget. </t> <t> The reference implementation keeps track of the total number of whole bits that have been processed by the decoder so far in the variable nbits_total, including the (possibly fractional) number of bits that are currently buffered, but not consumed, inside the range coder. nbits_total is initialized to 33 just after the initial range renormalization process completes (or equivalently, it can be initialized to 9 before the first renormalization). The extra two bits over the actual amount buffered by the range coder guarantees that it is an upper bound and that there is enough room for the encoder to terminate the stream. Each iteration through the range coder's renormalization loop increases nbits_total by 8. Reading raw bits increases nbits_total by the number of raw bits read. </t> <section anchor="ec_tell" title="ec_tell()"> <t> The whole number of bits buffered in rng may be estimated via l = ilog(rng). ec_tell() then becomes a simple matter of removing these bits from the total. It returns (nbits_total - l). </t> <t> In a newly initialized decoder, before any symbols have been read, this reports that 1 bit has been used. This is the bit reserved for termination of the encoder. </t> </section> <section anchor="ec_tell_frac" title="ec_tell_frac()"> <t> ec_tell_frac() estimates the number of bits buffered in rng to fractional precision. Since rng must be greater than 2**23 after renormalization, l must be at least 24. Let <figure align="center"> <artwork align="center"> <![CDATA[ r_Q15 = rng >> (l-16) , ]]></artwork> </figure> so that 32768 <= r_Q15 < 65536, an unsigned Q15 value representing the fractional part of rng. Then the following procedure can be used to add one bit of precision to l. First, update <figure align="center"> <artwork align="center"> <![CDATA[ r_Q15 = (r_Q15*r_Q15) >> 15 . ]]></artwork> </figure> Then add the 16th bit of r_Q15 to l via <figure align="center"> <artwork align="center"> <![CDATA[ l = 2*l + (r_Q15 >> 16) . ]]></artwork> </figure> Finally, if this bit was a 1, reduce r_Q15 by a factor of two via <figure align="center"> <artwork align="center"> <![CDATA[ r_Q15 = r_Q15 >> 1 , ]]></artwork> </figure> so that it once again lies in the range 32768 <= r_Q15 < 65536. </t> <t> This procedure is repeated three times to extend l to 1/8th bit precision. ec_tell_frac() then returns (nbits_total*8 - l). </t> </section> </section> </section> <section anchor='outline_decoder' title='SILK Decoder'> <t> The decoder's LP layer uses a modified version of the SILK codec (herein simply called "SILK"), which runs a decoded excitation signal through adaptive long-term and short-term prediction synthesis filters. It runs in NB, MB, and WB modes internally. When used in a hybrid frame in SWB or FB mode, the LP layer itself still only runs in WB mode. </t> <section title="SILK Decoder Modules"> <t> An overview of the decoder is given in <xref target="decoder_figure"/>. </t> <figure align="center" anchor="decoder_figure"> <artwork align="center"> <![CDATA[ +---------+ +------------+ -->| Range |--->| Decode |---------------------------+ 1 | Decoder | 2 | Parameters |----------+ 5 | +---------+ +------------+ 4 | | 3 | | | \/ \/ \/ +------------+ +------------+ +------------+ | Generate |-->| LTP |-->| LPC | | Excitation | | Synthesis | | Synthesis | +------------+ +------------+ +------------+ | +------------------------------------+ | 6 | +------------+ +------------+ +-->| Stereo |-->| Resampling |--> 8 | Unmixing | 7 | | 8 +------------+ +------------+ 1: Range encoded bitstream 2: Coded parameters 3: Pulses and gains 4: Pitch lags and LTP coefficients 5: LPC coefficients 6: Decoded signal (mono or mid-side stereo) 7: Unmixed signal (mono or left-right stereo) 8: Resampled signal ]]> </artwork> <postamble>Decoder block diagram.</postamble> </figure> <!--TODO: 3. needs to be fixed. a) "pulses" are only part of the excitation magnitude, and this distinction matters due to sign coding, and b) our actual decoder does not scale the excitation by the gains; instead it scales the filtered output--> <t> The decoder feeds the bitstream (1) to the range decoder from <xref target="range-decoder"/>, and then decodes the parameters in it (2) using the procedures detailed in Sections <xref format="counter" target="silk_header_bits"/> through <xref format="counter" target="silk_signs"/>. These parameters (3, 4, 5) are used to generate an excitation signal (see <xref target="silk_excitation_reconstruction"/>), which is fed to an optional long-term prediction (LTP) filter (voiced frames only, see <xref target="silk_ltp_synthesis"/>) and then a short-term prediction filter (see <xref target="silk_lpc_synthesis"/>), producing the decoded signal (6). For stereo streams, the mid-side representation is converted to separate left and right channels (7). The result is finally resampled to the desired output sample rate (e.g., 48 kHz) so that the resampled signal (8) can be mixed with the CELT layer. </t> </section> <!--TODO: Document mandated decoder resets--> <section anchor="silk_layer_organization" title="LP Layer Organization"> <t> Internally, the LP layer of a single Opus frame is composed of either a single 10 ms regular SILK frame or between one and three 20 ms regular SILK frames. A stereo Opus frame may double the number of regular SILK frames (up to a total of six), since it includes separate frames for a mid channel and, optionally, a side channel. Optional Low Bit-Rate Redundancy (LBRR) frames, which are reduced-bitrate encodings of previous SILK frames, may be included to aid in recovery from packet loss. If present, these appear before the regular SILK frames. They are in most respects identical to regular, active SILK frames, except that they are usually encoded with a lower bitrate. This draft uses "SILK frame" to refer to either one and "regular SILK frame" if it needs to draw a distinction between the two. </t> <t> Logically, each SILK frame is in turn composed of either two or four 5 ms subframes. Various parameters, such as the quantization gain of the excitation and the pitch lag and filter coefficients can vary on a subframe-by-subframe basis. Physically, the parameters for each subframe are interleaved in the bitstream, as described in the relevant sections for each parameter. </t> <t> All of these frames and subframes are decoded from the same range coder, with no padding between them. Thus packing multiple SILK frames in a single Opus frame saves, on average, half a byte per SILK frame. It also allows some parameters to be predicted from prior SILK frames in the same Opus frame, since this does not degrade packet loss robustness (beyond any penalty for merely using fewer, larger packets to store multiple frames). </t> <t> Stereo support in SILK uses a variant of mid-side coding, allowing a mono decoder to simply decode the mid channel. However, the data for the two channels is interleaved, so a mono decoder must still unpack the data for the side channel. It would be required to do so anyway for hybrid Opus frames, or to support decoding individual 20 ms frames. </t> <t> <xref target="silk_symbols"/> summarizes the overal grouping of the contents of the LP layer. Figures <xref format="counter" target="silk_mono_60ms_frame"/> and <xref format="counter" target="silk_stereo_60ms_frame"/> illustrate the ordering of the various SILK frames for a 60&nbps;ms Opus frame according to the rules described, for both mono and stereo, respectively. </t> <texttable anchor="silk_symbols"> <ttcol align="center">Symbol(s)</ttcol> <ttcol align="center">PDF(s)</ttcol> <ttcol align="center">Condition</ttcol> <c>VAD flags</c> <c>{1, 1}/2</c> <c/> <c>LBRR flag</c> <c>{1, 1}/2</c> <c/> <c>Per-frame LBRR flags</c> <c><xref target="silk_lbrr_flag_pdfs"/></c> <c><xref target="silk_lbrr_flags"/></c> <c>LBRR Frame(s)</c> <c><xref target="silk_frame"/></c> <c><xref target="silk_lbrr_flags"/></c> <c>Regular SILK Frame(s)</c> <c><xref target="silk_frame"/></c> <c/> <postamble> Organization of the SILK layer of an Opus frame. </postamble> </texttable> <figure align="center" anchor="silk_mono_60ms_frame" title="A 60 ms Mono Frame"> <artwork align="center"><![CDATA[ +---------------------------------+ | VAD Flags | +---------------------------------+ | LBRR Flag | +---------------------------------+ | Per-Frame LBRR Flags (Optional) | +---------------------------------+ | LBRR Frame 1 (Optional) | +---------------------------------+ | LBRR Frame 2 (Optional) | +---------------------------------+ | LBRR Frame 3 (Optional) | +---------------------------------+ | Regular SILK Frame 1 | +---------------------------------+ | Regular SILK Frame 2 | +---------------------------------+ | Regular SILK Frame 3 | +---------------------------------+ ]]></artwork> </figure> <figure align="center" anchor="silk_stereo_60ms_frame" title="A 60 ms Stereo Frame"> <artwork align="center"><![CDATA[ +---------------------------------------+ | Mid VAD Flags | +---------------------------------------+ | Mid LBRR Flag | +---------------------------------------+ | Side VAD Flags | +---------------------------------------+ | Side LBRR Flag | +---------------------------------------+ | Mid Per-Frame LBRR Flags (Optional) | +---------------------------------------+ | Side Per-Frame LBRR Flags (Optional) | +---------------------------------------+ | Mid LBRR Frame 1 (Optional) | +---------------------------------------+ | Side LBRR Frame 1 (Optional) | +---------------------------------------+ | Mid LBRR Frame 2 (Optional) | +---------------------------------------+ | Side LBRR Frame 2 (Optional) | +---------------------------------------+ | Mid LBRR Frame 3 (Optional) | +---------------------------------------+ | Side LBRR Frame 3 (Optional) | +---------------------------------------+ | Mid Regular SILK Frame 1 | +---------------------------------------+ | Side Regular SILK Frame 1 (Optional) | +---------------------------------------+ | Mid Regular SILK Frame 2 | +---------------------------------------+ | Side Regular SILK Frame 2 (Optional) | +---------------------------------------+ | Mid Regular SILK Frame 3 | +---------------------------------------+ | Side Regular SILK Frame 3 (Optional) | +---------------------------------------+ ]]></artwork> </figure> </section> <section anchor="silk_header_bits" title="Header Bits"> <t> The LP layer begins with two to eight header bits, decoded in silk_Decode() (dec_API.c). These consist of one Voice Activity Detection (VAD) bit per frame (up to 3), followed by a single flag indicating the presence of LBRR frames. For a stereo packet, these first flags correspond to the mid channel, and a second set of flags is included for the side channel. </t> <t> Because these are the first symbols decoded by the range coder and because they are coded as binary values with uniform probability, they can be extracted directly from the most significant bits of the first byte of compressed data. Thus, a receiver can determine if an Opus frame contains any active SILK frames without the overhead of using the range decoder. </t> </section> <section anchor="silk_lbrr_flags" title="Per-Frame LBRR Flags"> <t> For Opus frames longer than 20 ms, a set of LBRR flags is decoded for each channel that has its LBRR flag set. Each set contains one flag per 20 ms SILK frame. 40 ms Opus frames use the 2-frame LBRR flag PDF from <xref target="silk_lbrr_flag_pdfs"/>, and 60 ms Opus frames use the 3-frame LBRR flag PDF. For each channel, the resulting 2- or 3-bit integer contains the corresponding LBRR flag for each frame, packed in order from the LSb to the MSb. </t> <texttable anchor="silk_lbrr_flag_pdfs" title="LBRR Flag PDFs"> <ttcol>Frame Size</ttcol> <ttcol>PDF</ttcol> <c>40 ms</c> <c>{0, 53, 53, 150}/256</c> <c>60 ms</c> <c>{0, 41, 20, 29, 41, 15, 28, 82}/256</c> </texttable> <t> A 10 or 20 ms Opus frame does not contain any per-frame LBRR flags, as there may be at most one LBRR frame per channel. The global LBRR flag in the header bits (see <xref target="silk_header_bits"/>) is already sufficient to indicate the presence of that single LBRR frame. </t> </section> <section anchor="silk_lbrr_frames" title="LBRR Frames"> <t> The LBRR frames, if present, immediately follow, as indicated by the LBRR flags, and prior to any regular SILK frames. <xref target="silk_frame"/> describes their exact contents. LBRR frames do not include their own separate VAD flags. LBRR frames are only meant to be transmitted for active speech, thus all LBRR frames are treated as active. </t> <t> In a stereo Opus frame longer than 20 ms, although the per-frame LBRR flags for the mid channel are coded as a unit before the per-frame LBRR flags for the side channel, the LBRR frames themselves are interleaved. The decoder parses an LBRR frame for the mid channel of a given 20 ms interval (if present) and then immediately parses the corresponding LBRR frame for the side channel (if present), before proceeding to the next 20 ms interval. </t> </section> <section anchor="silk_regular_frames" title="Regular SILK Frames"> <t> The regular SILK frame(s) follow the LBRR frames (if any). <xref target="silk_frame"/> describes their contents, as well. Unlike the LBRR frames, a regular SILK frame is always coded for each time interval in an Opus frame, even if the corresponding VAD flag is unset. For stereo Opus frames longer than 20 ms, the regular mid and side SILK frames for each 20 ms interval are interleaved, just as with the LBRR frames. The side frame may be skipped by coding an appropriate flag, as detailed in <xref target="silk_mid_only_flag"/>. </t> </section> <section anchor="silk_frame" title="SILK Frame Contents"> <t> Each SILK frame includes a set of side information that encodes <list style="symbols"> <t>The frame type and quantization type (<xref target="silk_frame_type"/>),</t> <t>Quantization gains (<xref target="silk_gains"/>),</t> <t>Short-term prediction filter coefficients (<xref target="silk_nlsfs"/>),</t> <t>An LSF interpolation weight (<xref target="silk_nlsf_interpolation"/>),</t> <t> Long-term prediction filter lags and gains (<xref target="silk_ltp_params"/>), and </t> <t>A linear congruential generator (LCG) seed (<xref target="silk_seed"/>).</t> </list> The quantized excitation signal (see <xref target="silk_excitation"/>) follows these at the end of the frame. <xref target="silk_frame_symbols"/> details the overall organization of a SILK frame. </t> <texttable anchor="silk_frame_symbols"> <ttcol align="center">Symbol(s)</ttcol> <ttcol align="center">PDF(s)</ttcol> <ttcol align="center">Condition</ttcol> <c>Stereo Prediction Weights</c> <c><xref target="silk_stereo_pred_pdfs"/></c> <c><xref target="silk_stereo_pred"/></c> <c>Mid-Only Flag</c> <c><xref target="silk_mid_only_pdf"/></c> <c><xref target="silk_mid_only_flag"/></c> <c>Frame Type</c> <c><xref target="silk_frame_type"/></c> <c/> <c>Subframe Gains</c> <c><xref target="silk_gains"/></c> <c/> <c>Normalized LSF Stage 1 Index</c> <c><xref target="silk_nlsf_stage1_pdfs"/></c> <c/> <c>Normalized LSF Stage 2 Residual</c> <c><xref target="silk_nlsf_stage2"/></c> <c/> <c>Normalized LSF Interpolation Weight</c> <c><xref target="silk_nlsf_interp_pdf"/></c> <c><xref target="silk_nlsf_interpolation"/></c> <c>Primary Pitch Lag</c> <c><xref target="silk_ltp_lags"/></c> <c>Voiced frame</c> <c>Subframe Pitch Contour</c> <c><xref target="silk_pitch_contour_pdfs"/></c> <c>Voiced frame</c> <c>Periodicity Index</c> <c><xref target="silk_perindex_pdf"/></c> <c>Voiced frame</c> <c>LTP Filter</c> <c><xref target="silk_ltp_filter_pdfs"/></c> <c>Voiced frame</c> <c>LTP Scaling</c> <c><xref target="silk_ltp_scaling_pdf"/></c> <c><xref target="silk_ltp_scaling"/></c> <c>LCG Seed</c> <c><xref target="silk_seed_pdf"/></c> <c/> <c>Excitation Rate Level</c> <c><xref target="silk_rate_level_pdfs"/></c> <c/> <c>Excitation Pulse Counts</c> <c><xref target="silk_pulse_count_pdfs"/></c> <c/> <c>Excitation Pulse Locations</c> <c><xref target="silk_pulse_locations"/></c> <c>Non-zero pulse count</c> <c>Excitation LSb's</c> <c><xref target="silk_shell_lsb_pdf"/></c> <c><xref target="silk_pulse_counts"/></c> <c>Excitation Signs</c> <c><xref target="silk_sign_pdfs"/></c> <c><xref target="silk_signs"/></c> <postamble> Order of the symbols in an individual SILK frame. </postamble> </texttable> <section anchor="silk_stereo_pred" title="Stereo Prediction Weights"> <t> A SILK frame corresponding to the mid channel of a stereo Opus frame begins with a pair of side channel prediction weights, designed such that zeros indicate normal mid-side coupling. Since these weights can change on every frame, the first portion of each frame linearly interpolates between the previous weights and the current ones, using zeros for the previous weights if none are available. These prediction weights are never included in a mono Opus frame, and the previous weights are reset to zeros on any transition from a mono to stereo. They are also not included in an LBRR frame for the side channel, even if the LBRR flags indicate the corresponding mid channel was not coded. In that case, the previous weights are used, again substituting in zeros if no previous weights are available since the last decoder reset. </t> <t> To summarize, these weights are coded if and only if <list style="symbols"> <t>This is a stereo Opus frame (<xref target="toc_byte"/>), and</t> <t>The current SILK frame corresponds to the mid channel.</t> </list> </t> <t> The prediction weights are coded in three separate pieces, which are decoded by silk_stereo_decode_pred() (decode_stereo_pred.c). The first piece jointly codes the high-order part of a table index for both weights. The second piece codes the low-order part of each table index. The third piece codes an offset used to linearly interpolate between table indices. The details are as follows. </t> <t> Let n be an index decoded with the 25-element stage-1 PDF in <xref target="silk_stereo_pred_pdfs"/>. Then let i0 and i1 be indices decoded with the stage-2 and stage-3 PDFs in <xref target="silk_stereo_pred_pdfs"/>, respectively, and let i2 and i3 be two more indices decoded with the stage-2 and stage-3 PDFs, all in that order. </t> <texttable anchor="silk_stereo_pred_pdfs" title="Stereo Weight PDFs"> <ttcol align="left">Stage</ttcol> <ttcol align="left">PDF</ttcol> <c>Stage 1</c> <c>{7, 2, 1, 1, 1, 10, 24, 8, 1, 1, 3, 23, 92, 23, 3, 1, 1, 8, 24, 10, 1, 1, 1, 2, 7}/256</c> <c>Stage 2</c> <c>{85, 86, 85}/256</c> <c>Stage 3</c> <c>{51, 51, 52, 51, 51}/256</c> </texttable> <t> Then use n, i0, and i2 to form two table indices, wi0 and wi1, according to <figure align="center"> <artwork align="center"><![CDATA[ wi0 = i0 + 3*(n/5) wi1 = i2 + 3*(n%5) ]]></artwork> </figure> where the division is exact integer division. The range of these indices is 0 to 14, inclusive. Let w[i] be the i'th weight from <xref target="silk_stereo_weights_table"/>. Then the two prediction weights, w0_Q13 and w1_Q13, are <figure align="center"> <artwork align="center"><![CDATA[ w1_Q13 = w_Q13[wi1] + ((w_Q13[wi1+1] - w_Q13[wi1])*6554) >> 16)*(2*i3 + 1) w0_Q13 = w_Q13[wi0] + ((w_Q13[wi0+1] - w_Q13[wi0])*6554) >> 16)*(2*i1 + 1) - w1_Q13 ]]></artwork> </figure> N.b., w1_Q13 is computed first here, because w0_Q13 depends on it. </t> <texttable anchor="silk_stereo_weights_table" title="Stereo Weight Table"> <ttcol align="left">Index</ttcol> <ttcol align="right">Weight (Q13)</ttcol> <c>0</c> <c>-13732</c> <c>1</c> <c>-10050</c> <c>2</c> <c>-8266</c> <c>3</c> <c>-7526</c> <c>4</c> <c>-6500</c> <c>5</c> <c>-5000</c> <c>6</c> <c>-2950</c> <c>7</c> <c>-820</c> <c>8</c> <c>820</c> <c>9</c> <c>2950</c> <c>10</c> <c>5000</c> <c>11</c> <c>6500</c> <c>12</c> <c>7526</c> <c>13</c> <c>8266</c> <c>14</c> <c>10050</c> <c>15</c> <c>13732</c> </texttable> </section> <section anchor="silk_mid_only_flag" title="Mid-Only Flag"> <t> A flag appears after the stereo prediction weights that indicates if only the mid channel is coded for this time interval. It appears only when <list style="symbols"> <t>This is a stereo Opus frame (see <xref target="toc_byte"/>),</t> <t>The current SILK frame corresponds to the mid channel, and</t> <t>Either <list style="symbols"> <t>This is a regular SILK frame, or</t> <t> This is an LBRR frame where the corresponding LBRR flags (see <xref target="silk_header_bits"/> and <xref target="silk_lbrr_flags"/>) indicate the side channel is not coded. </t> </list> </t> </list> It is omitted when there are no stereo weights, and it is also omitted for an LBRR frame when the corresponding LBRR flags indicate the side channel is coded. When the flag is present, the decoder reads a single value using the PDF in <xref target="silk_mid_only_pdf"/>, as implemented in silk_stereo_decode_mid_only() (decode_stereo_pred.c). If the flag is set, then there is no corresponding SILK frame for the side channel, the entire decoding process for the side channel is skipped, and zeros are used during the stereo unmixing process<!--TODO: ref-->. As stated above, LBRR frames still include this flag when the LBRR flag indicates that the side channel is not coded. In that case, if this flag is zero (indicating that there should be a side channel), then Packet Loss Concealment (PLC, see <xref target="Packet Loss Concealment"/>) SHOULD be invoked to recover a side channel signal. </t> <texttable anchor="silk_mid_only_pdf" title="Mid-Only Flag PDF"> <ttcol align="left">PDF</ttcol> <c>{192, 64}/256</c> </texttable> </section> <section anchor="silk_frame_type" title="Frame Type"> <t> Each SILK frame contains a single "frame type" symbol that jointly codes the signal type and quantization offset type of the corresponding frame. If the current frame is a regular SILK frame whose VAD bit was not set (an "inactive" frame), then the frame type symbol takes on a value of either 0 or 1 and is decoded using the first PDF in <xref target="silk_frame_type_pdfs"/>. If the frame is an LBRR frame or a regular SILK frame whose VAD flag was set (an "active" frame), then the value of the symbol may range from 2 to 5, inclusive, and is decoded using the second PDF in <xref target="silk_frame_type_pdfs"/>. <xref target="silk_frame_type_table"/> translates between the value of the frame type symbol and the corresponding signal type and quantization offset type. </t> <texttable anchor="silk_frame_type_pdfs" title="Frame Type PDFs"> <ttcol>VAD Flag</ttcol> <ttcol>PDF</ttcol> <c>Inactive</c> <c>{26, 230, 0, 0, 0, 0}/256</c> <c>Active</c> <c>{0, 0, 24, 74, 148, 10}/256</c> </texttable> <texttable anchor="silk_frame_type_table" title="Signal Type and Quantization Offset Type from Frame Type"> <ttcol>Frame Type</ttcol> <ttcol>Signal Type</ttcol> <ttcol align="right">Quantization Offset Type</ttcol> <c>0</c> <c>Inactive</c> <c>Low</c> <c>1</c> <c>Inactive</c> <c>High</c> <c>2</c> <c>Unvoiced</c> <c>Low</c> <c>3</c> <c>Unvoiced</c> <c>High</c> <c>4</c> <c>Voiced</c> <c>Low</c> <c>5</c> <c>Voiced</c> <c>High</c> </texttable> </section> <section anchor="silk_gains" title="Subframe Gains"> <t> A separate quantization gain is coded for each 5 ms subframe. These gains control the step size between quantization levels of the excitation signal and, therefore, the quality of the reconstruction. They are independent of the pitch gains coded for voiced frames. The quantization gains are themselves uniformly quantized to 6 bits on a log scale, giving them a resolution of approximately 1.369 dB and a range of approximately 1.94 dB to 88.21 dB. </t> <t> The subframe gains are either coded independently, or relative to the gain from the most recent coded subframe in the same channel. Independent coding is used if and only if <list style="symbols"> <t> This is the first subframe in the current SILK frame, and </t> <t>Either <list style="symbols"> <t>This is the first LBRR frame for this channel in the current Opus frame,</t> <t> This is an LBRR frame where the LBRR flags (see <xref target="silk_header_bits"/> and <xref target="silk_lbrr_flags"/>) indicate the previous LBRR frame in the same channel is not coded, or </t> <t> This is the first regular SILK frame for this channel in the current Opus frame. </t> </list> </t> </list> </t> <t> There are a few subtle points here that may benefit from some clarification. The rules for uncoded LBRR frames are very different from the rules for regular SILK frames for the side channel of a stereo Opus frame. Both allow gaps in the sequence of coded frames for a channel, the former based on the LBRR flags, and the latter on the mid-only flag (from <xref target="silk_mid_only_flag"/>). LBRR frames do not use relative coding to predict across these gaps, while regular SILK frames in the side channel do. In particular, in a 60 ms stereo Opus frame, if the first and third regular SILK frames in the side channel are coded, but the second is not, the first subframe of the third frame is still coded relative to the last subframe in the first frame. In contrast, in a similar situation with LBRR frames, the first subframe of the third frame would use independent coding, even if the mid-only flag for the second frame was 0. </t> <t> In an independently coded subframe gain, the 3 most significant bits of the quantization gain are decoded using a PDF selected from <xref target="silk_independent_gain_msb_pdfs"/> based on the decoded signal type (see <xref target="silk_frame_type"/>). </t> <texttable anchor="silk_independent_gain_msb_pdfs" title="PDFs for Independent Quantization Gain MSb Coding"> <ttcol align="left">Signal Type</ttcol> <ttcol align="left">PDF</ttcol> <c>Inactive</c> <c>{32, 112, 68, 29, 12, 1, 1, 1}/256</c> <c>Unvoiced</c> <c>{2, 17, 45, 60, 62, 47, 19, 4}/256</c> <c>Voiced</c> <c>{1, 3, 26, 71, 94, 50, 9, 2}/256</c> </texttable> <t> The 3 least significant bits are decoded using a uniform PDF: </t> <texttable anchor="silk_independent_gain_lsb_pdf" title="PDF for Independent Quantization Gain LSb Coding"> <ttcol align="left">PDF</ttcol> <c>{32, 32, 32, 32, 32, 32, 32, 32}/256</c> </texttable> <t> For subframes which do not have an independent gain (including the first subframe of frames not listed as using independent coding above), the quantization gain is coded relative to the gain from the previous subframe (in the same channel). The PDF in <xref target="silk_delta_gain_pdf"/> yields a delta gain index between 0 and 40, inclusive. </t> <texttable anchor="silk_delta_gain_pdf" title="PDF for Delta Quantization Gain Coding"> <ttcol align="left">PDF</ttcol> <c>{6, 5, 11, 31, 132, 21, 8, 4, 3, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}/256</c> </texttable> <t> The following formula translates this index into a quantization gain for the current subframe using the gain from the previous subframe: </t> <figure align="center"> <artwork align="center"><![CDATA[ log_gain = min(max(2*gain_index - 16, previous_log_gain + gain_index - 4), 63) ]]></artwork> </figure> <t> silk_gains_dequant() (gain_quant.c) dequantizes the gain for the k'th subframe and converts it into a linear Q16 scale factor via <figure align="center"> <artwork align="center"><![CDATA[ gain_Q16[k] = silk_log2lin((0x1D1C71*log_gain>>16) + 2090) ]]></artwork> </figure> </t> <t> The function silk_log2lin() (log2lin.c) computes an approximation of 2**(inLog_Q7/128.0), where inLog_Q7 is its Q7 input. Let i = inLog_Q7>>7 be the integer part of inLogQ7 and f = inLog_Q7&127 be the fractional part. Then <figure align="center"> <artwork align="center"><![CDATA[ (128 + f + ((-174*f*(128-f))>>16)) << (i - 7) ]]></artwork> </figure> yields the approximate exponential. </t> </section> <section anchor="silk_nlsfs" title="Normalized Line Spectral Frequency (LSF) and Linear Predictive Coding (LPC) Coefficients"> <t> A set of normalized Line Spectral Frequency (LSF) coefficients follow the quantization gains in the bitstream, and represent the Linear Predictive Coding (LPC) coefficients for the current SILK frame. Once decoded, the normalized LSFs form an increasing list of Q15 values between 0 and 1. These represent the interleaved zeros on the unit circle between 0 and pi (hence "normalized") in the standard decomposition of the LPC filter into a symmetric part and an anti-symmetric part (P and Q in <xref target="silk_nlsf2lpc"/>). Because of non-linear effects in the decoding process, an implementation SHOULD match the fixed-point arithmetic described in this section exactly. An encoder SHOULD also use the same process. </t> <t> The normalized LSFs are coded using a two-stage vector quantizer (VQ) (<xref target="silk_nlsf_stage1"/> and <xref target="silk_nlsf_stage2"/>). NB and MB frames use an order-10 predictor, while WB frames use an order-16 predictor, and thus have different sets of tables. After reconstructing the normalized LSFs (<xref target="silk_nlsf_reconstruction"/>), the decoder runs them through a stabilization process (<xref target="silk_nlsf_stabilization"/>), interpolates them between frames (<xref target="silk_nlsf_interpolation"/>), converts them back into LPC coefficients (<xref target="silk_nlsf2lpc"/>), and then runs them through further processes to limit the range of the coefficients (<xref target="silk_lpc_range_limit"/>) and the gain of the filter (<xref target="silk_lpc_gain_limit"/>). All of this is necessary to ensure the reconstruction process is stable. </t> <section anchor="silk_nlsf_stage1" title="Stage 1 Normalized LSF Decoding"> <t> The first VQ stage uses a 32-element codebook, coded with one of the PDFs in <xref target="silk_nlsf_stage1_pdfs"/>, depending on the audio bandwidth and the signal type of the current SILK frame. This yields a single index, I1, for the entire frame. This indexes an element in a coarse codebook, selects the PDFs for the second stage of the VQ, and selects the prediction weights used to remove intra-frame redundancy from the second stage. The actual codebook elements are listed in <xref target="silk_nlsf_nbmb_codebook"/> and <xref target="silk_nlsf_wb_codebook"/>, but they are not needed until the last stages of reconstructing the LSF coefficients. </t> <texttable anchor="silk_nlsf_stage1_pdfs" title="PDFs for Normalized LSF Index Stage-1 Decoding"> <ttcol align="left">Audio Bandwidth</ttcol> <ttcol align="left">Signal Type</ttcol> <ttcol align="left">PDF</ttcol> <c>NB or MB</c> <c>Inactive or unvoiced</c> <c> {44, 34, 30, 19, 21, 12, 11, 3, 3, 2, 16, 2, 2, 1, 5, 2, 1, 3, 3, 1, 1, 2, 2, 2, 3, 1, 9, 9, 2, 7, 2, 1}/256 </c> <c>NB or MB</c> <c>Voiced</c> <c> {1, 10, 1, 8, 3, 8, 8, 14, 13, 14, 1, 14, 12, 13, 11, 11, 12, 11, 10, 10, 11, 8, 9, 8, 7, 8, 1, 1, 6, 1, 6, 5}/256 </c> <c>WB</c> <c>Inactive or unvoiced</c> <c> {31, 21, 3, 17, 1, 8, 17, 4, 1, 18, 16, 4, 2, 3, 1, 10, 1, 3, 16, 11, 16, 2, 2, 3, 2, 11, 1, 4, 9, 8, 7, 3}/256 </c> <c>WB</c> <c>Voiced</c> <c> {1, 4, 16, 5, 18, 11, 5, 14, 15, 1, 3, 12, 13, 14, 14, 6, 14, 12, 2, 6, 1, 12, 12, 11, 10, 3, 10, 5, 1, 1, 1, 3}/256 </c> </texttable> </section> <section anchor="silk_nlsf_stage2" title="Stage 2 Normalized LSF Decoding"> <t> A total of 16 PDFs are available for the LSF residual in the second stage: the 8 (a...h) for NB and MB frames given in <xref target="silk_nlsf_stage2_nbmb_pdfs"/>, and the 8 (i...p) for WB frames given in <xref target="silk_nlsf_stage2_wb_pdfs"/>. Which PDF is used for which coefficient is driven by the index, I1, decoded in the first stage. <xref target="silk_nlsf_nbmb_stage2_cb_sel"/> lists the letter of the corresponding PDF for each normalized LSF coefficient for NB and MB, and <xref target="silk_nlsf_wb_stage2_cb_sel"/> lists the same information for WB. </t> <texttable anchor="silk_nlsf_stage2_nbmb_pdfs" title="PDFs for NB/MB Normalized LSF Index Stage-2 Decoding"> <ttcol align="left">Codebook</ttcol> <ttcol align="left">PDF</ttcol> <c>a</c> <c>{1, 1, 1, 15, 224, 11, 1, 1, 1}/256</c> <c>b</c> <c>{1, 1, 2, 34, 183, 32, 1, 1, 1}/256</c> <c>c</c> <c>{1, 1, 4, 42, 149, 55, 2, 1, 1}/256</c> <c>d</c> <c>{1, 1, 8, 52, 123, 61, 8, 1, 1}/256</c> <c>e</c> <c>{1, 3, 16, 53, 101, 74, 6, 1, 1}/256</c> <c>f</c> <c>{1, 3, 17, 55, 90, 73, 15, 1, 1}/256</c> <c>g</c> <c>{1, 7, 24, 53, 74, 67, 26, 3, 1}/256</c> <c>h</c> <c>{1, 1, 18, 63, 78, 58, 30, 6, 1}/256</c> </texttable> <texttable anchor="silk_nlsf_stage2_wb_pdfs" title="PDFs for WB Normalized LSF Index Stage-2 Decoding"> <ttcol align="left">Codebook</ttcol> <ttcol align="left">PDF</ttcol> <c>i</c> <c>{1, 1, 1, 9, 232, 9, 1, 1, 1}/256</c> <c>j</c> <c>{1, 1, 2, 28, 186, 35, 1, 1, 1}/256</c> <c>k</c> <c>{1, 1, 3, 42, 152, 53, 2, 1, 1}/256</c> <c>l</c> <c>{1, 1, 10, 49, 126, 65, 2, 1, 1}/256</c> <c>m</c> <c>{1, 4, 19, 48, 100, 77, 5, 1, 1}/256</c> <c>n</c> <c>{1, 1, 14, 54, 100, 72, 12, 1, 1}/256</c> <c>o</c> <c>{1, 1, 15, 61, 87, 61, 25, 4, 1}/256</c> <c>p</c> <c>{1, 7, 21, 50, 77, 81, 17, 1, 1}/256</c> </texttable> <texttable anchor="silk_nlsf_nbmb_stage2_cb_sel" title="Codebook Selection for NB/MB Normalized LSF Index Stage 2 Decoding"> <ttcol>I1</ttcol> <ttcol>Coefficient</ttcol> <c/> <c><spanx style="vbare">0 1 2 3 4 5 6 7 8 9</spanx></c> <c> 0</c> <c><spanx style="vbare">a a a a a a a a a a</spanx></c> <c> 1</c> <c><spanx style="vbare">b d b c c b c b b b</spanx></c> <c> 2</c> <c><spanx style="vbare">c b b b b b b b b b</spanx></c> <c> 3</c> <c><spanx style="vbare">b c c c c b c b b b</spanx></c> <c> 4</c> <c><spanx style="vbare">c d d d d c c c c c</spanx></c> <c> 5</c> <c><spanx style="vbare">a f d d c c c c b b</spanx></c> <c> g</c> <c><spanx style="vbare">a c c c c c c c c b</spanx></c> <c> 7</c> <c><spanx style="vbare">c d g e e e f e f f</spanx></c> <c> 8</c> <c><spanx style="vbare">c e f f e f e g e e</spanx></c> <c> 9</c> <c><spanx style="vbare">c e e h e f e f f e</spanx></c> <c>10</c> <c><spanx style="vbare">e d d d c d c c c c</spanx></c> <c>11</c> <c><spanx style="vbare">b f f g e f e f f f</spanx></c> <c>12</c> <c><spanx style="vbare">c h e g f f f f f f</spanx></c> <c>13</c> <c><spanx style="vbare">c h f f f f f g f e</spanx></c> <c>14</c> <c><spanx style="vbare">d d f e e f e f e e</spanx></c> <c>15</c> <c><spanx style="vbare">c d d f f e e e e e</spanx></c> <c>16</c> <c><spanx style="vbare">c e e g e f e f f f</spanx></c> <c>17</c> <c><spanx style="vbare">c f e g f f f e f e</spanx></c> <c>18</c> <c><spanx style="vbare">c h e f e f e f f f</spanx></c> <c>19</c> <c><spanx style="vbare">c f e g h g f g f e</spanx></c> <c>20</c> <c><spanx style="vbare">d g h e g f f g e f</spanx></c> <c>21</c> <c><spanx style="vbare">c h g e e e f e f f</spanx></c> <c>22</c> <c><spanx style="vbare">e f f e g g f g f e</spanx></c> <c>23</c> <c><spanx style="vbare">c f f g f g e g e e</spanx></c> <c>24</c> <c><spanx style="vbare">e f f f d h e f f e</spanx></c> <c>25</c> <c><spanx style="vbare">c d e f f g e f f e</spanx></c> <c>26</c> <c><spanx style="vbare">c d c d d e c d d d</spanx></c> <c>27</c> <c><spanx style="vbare">b b c c c c c d c c</spanx></c> <c>28</c> <c><spanx style="vbare">e f f g g g f g e f</spanx></c> <c>29</c> <c><spanx style="vbare">d f f e e e e d d c</spanx></c> <c>30</c> <c><spanx style="vbare">c f d h f f e e f e</spanx></c> <c>31</c> <c><spanx style="vbare">e e f e f g f g f e</spanx></c> </texttable> <texttable anchor="silk_nlsf_wb_stage2_cb_sel" title="Codebook Selection for WB Normalized LSF Index Stage 2 Decoding"> <ttcol>I1</ttcol> <ttcol>Coefficient</ttcol> <c/> <c><spanx style="vbare">0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15</spanx></c> <c> 0</c> <c><spanx style="vbare">i i i i i i i i i i i i i i i i</spanx></c> <c> 1</c> <c><spanx style="vbare">k l l l l l k k k k k j j j i l</spanx></c> <c> 2</c> <c><spanx style="vbare">k n n l p m m n k n m n n m l l</spanx></c> <c> 3</c> <c><spanx style="vbare">i k j k k j j j j j i i i i i j</spanx></c> <c> 4</c> <c><spanx style="vbare">i o n m o m p n m m m n n m m l</spanx></c> <c> 5</c> <c><spanx style="vbare">i l n n m l l n l l l l l l k m</spanx></c> <c> 6</c> <c><spanx style="vbare">i i i i i i i i i i i i i i i i</spanx></c> <c> 7</c> <c><spanx style="vbare">i k o l p k n l m n n m l l k l</spanx></c> <c> 8</c> <c><spanx style="vbare">i o k o o m n m o n m m n l l l</spanx></c> <c> 9</c> <c><spanx style="vbare">k j i i i i i i i i i i i i i i</spanx></c> <c>j0</c> <c><spanx style="vbare">i j i i i i i i i i i i i i i j</spanx></c> <c>11</c> <c><spanx style="vbare">k k l m n l l l l l l l k k j l</spanx></c> <c>12</c> <c><spanx style="vbare">k k l l m l l l l l l l l k j l</spanx></c> <c>13</c> <c><spanx style="vbare">l m m m o m m n l n m m n m l m</spanx></c> <c>14</c> <c><spanx style="vbare">i o m n m p n k o n p m m l n l</spanx></c> <c>15</c> <c><spanx style="vbare">i j i j j j j j j j i i i i j i</spanx></c> <c>16</c> <c><spanx style="vbare">j o n p n m n l m n m m m l l m</spanx></c> <c>17</c> <c><spanx style="vbare">j l l m m l l n k l l n n n l m</spanx></c> <c>18</c> <c><spanx style="vbare">k l l k k k l k j k j k j j j m</spanx></c> <c>19</c> <c><spanx style="vbare">i k l n l l k k k j j i i i i i</spanx></c> <c>20</c> <c><spanx style="vbare">l m l n l l k k j j j j j k k m</spanx></c> <c>21</c> <c><spanx style="vbare">k o l p p m n m n l n l l k l l</spanx></c> <c>22</c> <c><spanx style="vbare">k l n o o l n l m m l l l l k m</spanx></c> <c>23</c> <c><spanx style="vbare">j l l m m m m l n n n l j j j j</spanx></c> <c>24</c> <c><spanx style="vbare">k n l o o m p m m n l m m l l l</spanx></c> <c>25</c> <c><spanx style="vbare">i o j j i i i i i i i i i i i i</spanx></c> <c>26</c> <c><spanx style="vbare">i o o l n k n n l m m p p m m m</spanx></c> <c>27</c> <c><spanx style="vbare">l l p l n m l l l k k l l l k l</spanx></c> <c>28</c> <c><spanx style="vbare">i i j i i i k j k j j k k k j j</spanx></c> <c>29</c> <c><spanx style="vbare">i l k n l l k l k j i i j i i j</spanx></c> <c>30</c> <c><spanx style="vbare">l n n m p n l l k l k k j i j i</spanx></c> <c>31</c> <c><spanx style="vbare">k l n l m l l l k j k o m i i i</spanx></c> </texttable> <t> Decoding the second stage residual proceeds as follows. For each coefficient, the decoder reads a symbol using the PDF corresponding to I1 from either <xref target="silk_nlsf_nbmb_stage2_cb_sel"/> or <xref target="silk_nlsf_wb_stage2_cb_sel"/>, and subtracts 4 from the result to give an index in the range -4 to 4, inclusive. If the index is either -4 or 4, it reads a second symbol using the PDF in <xref target="silk_nlsf_ext_pdf"/>, and adds the value of this second symbol to the index, using the same sign. This gives the index, I2[k], a total range of -10 to 10, inclusive. </t> <texttable anchor="silk_nlsf_ext_pdf" title="PDF for Normalized LSF Index Extension Decoding"> <ttcol align="left">PDF</ttcol> <c>{156, 60, 24, 9, 4, 2, 1}/256</c> </texttable> <t> The decoded indices from both stages are translated back into normalized LSF coefficients in silk_NLSF_decode() (NLSF_decode.c). The stage-2 indices represent residuals after both the first stage of the VQ and a separate backwards-prediction step. The backwards prediction process in the encoder subtracts a prediction from each residual formed by a multiple of the coefficient that follows it. The decoder must undo this process. <xref target="silk_nlsf_pred_weights"/> contains lists of prediction weights for each coefficient. There are two lists for NB and MB, and another two lists for WB, giving two possible prediction weights for each coefficient. </t> <texttable anchor="silk_nlsf_pred_weights" title="Prediction Weights for Normalized LSF Decoding"> <ttcol align="left">Coefficient</ttcol> <ttcol align="right">A</ttcol> <ttcol align="right">B</ttcol> <ttcol align="right">C</ttcol> <ttcol align="right">D</ttcol> <c>0</c> <c>179</c> <c>116</c> <c>175</c> <c>68</c> <c>1</c> <c>138</c> <c>67</c> <c>148</c> <c>62</c> <c>2</c> <c>140</c> <c>82</c> <c>160</c> <c>66</c> <c>3</c> <c>148</c> <c>59</c> <c>176</c> <c>60</c> <c>4</c> <c>151</c> <c>92</c> <c>178</c> <c>72</c> <c>5</c> <c>149</c> <c>72</c> <c>173</c> <c>117</c> <c>6</c> <c>153</c> <c>100</c> <c>174</c> <c>85</c> <c>7</c> <c>151</c> <c>89</c> <c>164</c> <c>90</c> <c>8</c> <c>163</c> <c>92</c> <c>177</c> <c>118</c> <c>9</c> <c/> <c/> <c>174</c> <c>136</c> <c>10</c> <c/> <c/> <c>196</c> <c>151</c> <c>11</c> <c/> <c/> <c>182</c> <c>142</c> <c>12</c> <c/> <c/> <c>198</c> <c>160</c> <c>13</c> <c/> <c/> <c>192</c> <c>142</c> <c>14</c> <c/> <c/> <c>182</c> <c>155</c> </texttable> <t> The prediction is undone using the procedure implemented in silk_NLSF_residual_dequant() (NLSF_decode.c), which is as follows. Each coefficient selects its prediction weight from one of the two lists based on the stage-1 index, I1. <xref target="silk_nlsf_nbmb_weight_sel"/> gives the selections for each coefficient for NB and MB, and <xref target="silk_nlsf_wb_weight_sel"/> gives the selections for WB. Let d_LPC be the order of the codebook, i.e., 10 for NB and MB, and 16 for WB, and let pred_Q8[k] be the weight for the k'th coefficient selected by this process for 0 <= k < d_LPC-1. Then, the stage-2 residual for each coefficient is computed via <figure align="center"> <artwork align="center"><![CDATA[ res_Q10[k] = (k+1 < d_LPC ? (res_Q10[k+1]*pred_Q8[k])>>8 : 0) + ((((I2[k]<<10) + sign(I2[k])*102)*qstep)>>16) , ]]></artwork> </figure> where qstep is the Q16 quantization step size, which is 11796 for NB and MB and 9830 for WB (representing step sizes of approximately 0.18 and 0.15, respectively). </t> <texttable anchor="silk_nlsf_nbmb_weight_sel" title="Prediction Weight Selection for NB/MB Normalized LSF Decoding"> <ttcol>I1</ttcol> <ttcol>Coefficient</ttcol> <c/> <c><spanx style="vbare">0 1 2 3 4 5 6 7 8</spanx></c> <c> 0</c> <c><spanx style="vbare">A B A A A A A A A</spanx></c> <c> 1</c> <c><spanx style="vbare">B A A A A A A A A</spanx></c> <c> 2</c> <c><spanx style="vbare">A A A A A A A A A</spanx></c> <c> 3</c> <c><spanx style="vbare">B B B A A A A B A</spanx></c> <c> 4</c> <c><spanx style="vbare">A B A A A A A A A</spanx></c> <c> 5</c> <c><spanx style="vbare">A B A A A A A A A</spanx></c> <c> 6</c> <c><spanx style="vbare">B A B B A A A B A</spanx></c> <c> 7</c> <c><spanx style="vbare">A B B A A B B A A</spanx></c> <c> 8</c> <c><spanx style="vbare">A A B B A B A B B</spanx></c> <c> 9</c> <c><spanx style="vbare">A A B B A A B B B</spanx></c> <c>10</c> <c><spanx style="vbare">A A A A A A A A A</spanx></c> <c>11</c> <c><spanx style="vbare">A B A B B B B B A</spanx></c> <c>12</c> <c><spanx style="vbare">A B A B B B B B A</spanx></c> <c>13</c> <c><spanx style="vbare">A B B B B B B B A</spanx></c> <c>14</c> <c><spanx style="vbare">B A B B A B B B B</spanx></c> <c>15</c> <c><spanx style="vbare">A B B B B B A B A</spanx></c> <c>16</c> <c><spanx style="vbare">A A B B A B A B A</spanx></c> <c>17</c> <c><spanx style="vbare">A A B B B A B B B</spanx></c> <c>18</c> <c><spanx style="vbare">A B B A A B B B A</spanx></c> <c>19</c> <c><spanx style="vbare">A A A B B B A B A</spanx></c> <c>20</c> <c><spanx style="vbare">A B B A A B A B A</spanx></c> <c>21</c> <c><spanx style="vbare">A B B A A A B B A</spanx></c> <c>22</c> <c><spanx style="vbare">A A A A A B B B B</spanx></c> <c>23</c> <c><spanx style="vbare">A A B B A A A B B</spanx></c> <c>24</c> <c><spanx style="vbare">A A A B A B B B B</spanx></c> <c>25</c> <c><spanx style="vbare">A B B B B B B B A</spanx></c> <c>26</c> <c><spanx style="vbare">A A A A A A A A A</spanx></c> <c>27</c> <c><spanx style="vbare">A A A A A A A A A</spanx></c> <c>28</c> <c><spanx style="vbare">A A B A B B A B A</spanx></c> <c>29</c> <c><spanx style="vbare">A A A B A A A A A</spanx></c> <c>30</c> <c><spanx style="vbare">A A A B B A B A B</spanx></c> <c>31</c> <c><spanx style="vbare">B A B B A B B B B</spanx></c> </texttable> <texttable anchor="silk_nlsf_wb_weight_sel" title="Prediction Weight Selection for WB Normalized LSF Decoding"> <ttcol>I1</ttcol> <ttcol>Coefficient</ttcol> <c/> <c><spanx style="vbare">0 1 2 3 4 5 6 7 8 9 10 11 12 13 14</spanx></c> <c> 0</c> <c><spanx style="vbare">C C C C C C C C C C C C C C D</spanx></c> <c> 1</c> <c><spanx style="vbare">C C C C C C C C C C C C C C C</spanx></c> <c> 2</c> <c><spanx style="vbare">C C D C C D D D C D D D D C C</spanx></c> <c> 3</c> <c><spanx style="vbare">C C C C C C C C C C C C D C C</spanx></c> <c> 4</c> <c><spanx style="vbare">C D D C D C D D C D D D D D C</spanx></c> <c> 5</c> <c><spanx style="vbare">C D C C C C C C C C C C C C C</spanx></c> <c> 6</c> <c><spanx style="vbare">D C C C C C C C C C C D C D C</spanx></c> <c> 7</c> <c><spanx style="vbare">C D D C C C D C D D D C D C D</spanx></c> <c> 8</c> <c><spanx style="vbare">C D C D D C D C D C D D D D D</spanx></c> <c> 9</c> <c><spanx style="vbare">C C C C C C C C C C C C C C D</spanx></c> <c>10</c> <c><spanx style="vbare">C D C C C C C C C C C C C C C</spanx></c> <c>11</c> <c><spanx style="vbare">C C D C D D D D D D D C D C C</spanx></c> <c>12</c> <c><spanx style="vbare">C C D C C D C D C D C C D C C</spanx></c> <c>13</c> <c><spanx style="vbare">C C C C D D C D C D D D D C C</spanx></c> <c>14</c> <c><spanx style="vbare">C D C C C D D C D D D C D D D</spanx></c> <c>15</c> <c><spanx style="vbare">C C D D C C C C C C C C D D C</spanx></c> <c>16</c> <c><spanx style="vbare">C D D C D C D D D D D C D C C</spanx></c> <c>17</c> <c><spanx style="vbare">C C D C C C C D C C D D D C C</spanx></c> <c>18</c> <c><spanx style="vbare">C C C C C C C C C C C C C C D</spanx></c> <c>19</c> <c><spanx style="vbare">C C C C C C C C C C C C D C C</spanx></c> <c>20</c> <c><spanx style="vbare">C C C C C C C C C C C C C C C</spanx></c> <c>21</c> <c><spanx style="vbare">C D C D C D D C D C D C D D C</spanx></c> <c>22</c> <c><spanx style="vbare">C C D D D D C D D C C D D C C</spanx></c> <c>23</c> <c><spanx style="vbare">C D D C D C D C D C C C C D C</spanx></c> <c>24</c> <c><spanx style="vbare">C C C D D C D C D D D D D D D</spanx></c> <c>25</c> <c><spanx style="vbare">C C C C C C C C C C C C C C D</spanx></c> <c>26</c> <c><spanx style="vbare">C D D C C C D D C C D D D D D</spanx></c> <c>27</c> <c><spanx style="vbare">C C C C C D C D D D D C D D D</spanx></c> <c>28</c> <c><spanx style="vbare">C C C C C C C C C C C C C C D</spanx></c> <c>29</c> <c><spanx style="vbare">C C C C C C C C C C C C C C D</spanx></c> <c>30</c> <c><spanx style="vbare">D C C C C C C C C C C D C C C</spanx></c> <c>31</c> <c><spanx style="vbare">C C D C C D D D C C D C C D C</spanx></c> </texttable> </section> <section anchor="silk_nlsf_reconstruction" title="Reconstructing the Normalized LSF Coefficients"> <t> Once the stage-1 index I1 and the stage-2 residual res_Q10[] have been decoded, the final normalized LSF coefficients can be reconstructed. </t> <t> The spectral distortion introduced by the quantization of each LSF coefficient varies, so the stage-2 residual is weighted accordingly, using the low-complexity Inverse Harmonic Mean Weighting (IHMW) function proposed in <xref target="laroia-icassp"/>. The weights are derived directly from the stage-1 codebook vector. Let cb1_Q8[k] be the k'th entry of the stage-1 codebook vector from <xref target="silk_nlsf_nbmb_codebook"/> or <xref target="silk_nlsf_wb_codebook"/>. Then for 0 <= k < d_LPC the following expression computes the square of the weight as a Q18 value: <figure align="center"> <artwork align="center"> <![CDATA[ w2_Q18[k] = (1024/(cb1_Q8[k] - cb1_Q8[k-1]) + 1024/(cb1_Q8[k+1] - cb1_Q8[k])) << 16 , ]]> </artwork> </figure> where cb1_Q8[-1] = 0 and cb1_Q8[d_LPC] = 256, and the division is exact integer division. This is reduced to an unsquared, Q9 value using the following square-root approximation: <figure align="center"> <artwork align="center"><![CDATA[ i = ilog(w2_Q18[k]) f = (w2_Q18[k]>>(i-8)) & 127 y = ((i&1) ? 32768 : 46214) >> ((32-i)>>1) w_Q9[k] = y + ((213*f*y)>>16) ]]></artwork> </figure> The cb1_Q8[] vector completely determines these weights, and they may be tabulated and stored as 13-bit unsigned values (with a range of 1819 to 5227, inclusive) to avoid computing them when decoding. The reference implementation already requires code to compute these weights on unquantized coefficients in the encoder, in silk_NLSF_VQ_weights_laroia() (NLSF_VQ_weights_laroia.c) and its callers, so it reuses that code in the decoder instead of using a pre-computed table to reduce the amount of ROM required. </t> <texttable anchor="silk_nlsf_nbmb_codebook" title="Codebook Vectors for NB/MB Normalized LSF Stage 1 Decoding"> <ttcol>I1</ttcol> <ttcol>Codebook (Q8)</ttcol> <c/> <c><spanx style="vbare"> 0 1 2 3 4 5 6 7 8 9</spanx></c> <c>0</c> <c><spanx style="vbare">12 35 60 83 108 132 157 180 206 228</spanx></c> <c>1</c> <c><spanx style="vbare">15 32 55 77 101 125 151 175 201 225</spanx></c> <c>2</c> <c><spanx style="vbare">19 42 66 89 114 137 162 184 209 230</spanx></c> <c>3</c> <c><spanx style="vbare">12 25 50 72 97 120 147 172 200 223</spanx></c> <c>4</c> <c><spanx style="vbare">26 44 69 90 114 135 159 180 205 225</spanx></c> <c>5</c> <c><spanx style="vbare">13 22 53 80 106 130 156 180 205 228</spanx></c> <c>6</c> <c><spanx style="vbare">15 25 44 64 90 115 142 168 196 222</spanx></c> <c>7</c> <c><spanx style="vbare">19 24 62 82 100 120 145 168 190 214</spanx></c> <c>8</c> <c><spanx style="vbare">22 31 50 79 103 120 151 170 203 227</spanx></c> <c>9</c> <c><spanx style="vbare">21 29 45 65 106 124 150 171 196 224</spanx></c> <c>10</c> <c><spanx style="vbare">30 49 75 97 121 142 165 186 209 229</spanx></c> <c>11</c> <c><spanx style="vbare">19 25 52 70 93 116 143 166 192 219</spanx></c> <c>12</c> <c><spanx style="vbare">26 34 62 75 97 118 145 167 194 217</spanx></c> <c>13</c> <c><spanx style="vbare">25 33 56 70 91 113 143 165 196 223</spanx></c> <c>14</c> <c><spanx style="vbare">21 34 51 72 97 117 145 171 196 222</spanx></c> <c>15</c> <c><spanx style="vbare">20 29 50 67 90 117 144 168 197 221</spanx></c> <c>16</c> <c><spanx style="vbare">22 31 48 66 95 117 146 168 196 222</spanx></c> <c>17</c> <c><spanx style="vbare">24 33 51 77 116 134 158 180 200 224</spanx></c> <c>18</c> <c><spanx style="vbare">21 28 70 87 106 124 149 170 194 217</spanx></c> <c>19</c> <c><spanx style="vbare">26 33 53 64 83 117 152 173 204 225</spanx></c> <c>20</c> <c><spanx style="vbare">27 34 65 95 108 129 155 174 210 225</spanx></c> <c>21</c> <c><spanx style="vbare">20 26 72 99 113 131 154 176 200 219</spanx></c> <c>22</c> <c><spanx style="vbare">34 43 61 78 93 114 155 177 205 229</spanx></c> <c>23</c> <c><spanx style="vbare">23 29 54 97 124 138 163 179 209 229</spanx></c> <c>24</c> <c><spanx style="vbare">30 38 56 89 118 129 158 178 200 231</spanx></c> <c>25</c> <c><spanx style="vbare">21 29 49 63 85 111 142 163 193 222</spanx></c> <c>26</c> <c><spanx style="vbare">27 48 77 103 133 158 179 196 215 232</spanx></c> <c>27</c> <c><spanx style="vbare">29 47 74 99 124 151 176 198 220 237</spanx></c> <c>28</c> <c><spanx style="vbare">33 42 61 76 93 121 155 174 207 225</spanx></c> <c>29</c> <c><spanx style="vbare">29 53 87 112 136 154 170 188 208 227</spanx></c> <c>30</c> <c><spanx style="vbare">24 30 52 84 131 150 166 186 203 229</spanx></c> <c>31</c> <c><spanx style="vbare">37 48 64 84 104 118 156 177 201 230</spanx></c> </texttable> <texttable anchor="silk_nlsf_wb_codebook" title="Codebook Vectors for WB Normalized LSF Stage 1 Decoding"> <ttcol>I1</ttcol> <ttcol>Codebook (Q8)</ttcol> <c/> <c><spanx style="vbare"> 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15</spanx></c> <c>0</c> <c><spanx style="vbare"> 7 23 38 54 69 85 100 116 131 147 162 178 193 208 223 239</spanx></c> <c>1</c> <c><spanx style="vbare">13 25 41 55 69 83 98 112 127 142 157 171 187 203 220 236</spanx></c> <c>2</c> <c><spanx style="vbare">15 21 34 51 61 78 92 106 126 136 152 167 185 205 225 240</spanx></c> <c>3</c> <c><spanx style="vbare">10 21 36 50 63 79 95 110 126 141 157 173 189 205 221 237</spanx></c> <c>4</c> <c><spanx style="vbare">17 20 37 51 59 78 89 107 123 134 150 164 184 205 224 240</spanx></c> <c>5</c> <c><spanx style="vbare">10 15 32 51 67 81 96 112 129 142 158 173 189 204 220 236</spanx></c> <c>6</c> <c><spanx style="vbare"> 8 21 37 51 65 79 98 113 126 138 155 168 179 192 209 218</spanx></c> <c>7</c> <c><spanx style="vbare">12 15 34 55 63 78 87 108 118 131 148 167 185 203 219 236</spanx></c> <c>8</c> <c><spanx style="vbare">16 19 32 36 56 79 91 108 118 136 154 171 186 204 220 237</spanx></c> <c>9</c> <c><spanx style="vbare">11 28 43 58 74 89 105 120 135 150 165 180 196 211 226 241</spanx></c> <c>10</c> <c><spanx style="vbare"> 6 16 33 46 60 75 92 107 123 137 156 169 185 199 214 225</spanx></c> <c>11</c> <c><spanx style="vbare">11 19 30 44 57 74 89 105 121 135 152 169 186 202 218 234</spanx></c> <c>12</c> <c><spanx style="vbare">12 19 29 46 57 71 88 100 120 132 148 165 182 199 216 233</spanx></c> <c>13</c> <c><spanx style="vbare">17 23 35 46 56 77 92 106 123 134 152 167 185 204 222 237</spanx></c> <c>14</c> <c><spanx style="vbare">14 17 45 53 63 75 89 107 115 132 151 171 188 206 221 240</spanx></c> <c>15</c> <c><spanx style="vbare"> 9 16 29 40 56 71 88 103 119 137 154 171 189 205 222 237</spanx></c> <c>16</c> <c><spanx style="vbare">16 19 36 48 57 76 87 105 118 132 150 167 185 202 218 236</spanx></c> <c>17</c> <c><spanx style="vbare">12 17 29 54 71 81 94 104 126 136 149 164 182 201 221 237</spanx></c> <c>18</c> <c><spanx style="vbare">15 28 47 62 79 97 115 129 142 155 168 180 194 208 223 238</spanx></c> <c>19</c> <c><spanx style="vbare"> 8 14 30 45 62 78 94 111 127 143 159 175 192 207 223 239</spanx></c> <c>20</c> <c><spanx style="vbare">17 30 49 62 79 92 107 119 132 145 160 174 190 204 220 235</spanx></c> <c>21</c> <c><spanx style="vbare">14 19 36 45 61 76 91 108 121 138 154 172 189 205 222 238</spanx></c> <c>22</c> <c><spanx style="vbare">12 18 31 45 60 76 91 107 123 138 154 171 187 204 221 236</spanx></c> <c>23</c> <c><spanx style="vbare">13 17 31 43 53 70 83 103 114 131 149 167 185 203 220 237</spanx></c> <c>24</c> <c><spanx style="vbare">17 22 35 42 58 78 93 110 125 139 155 170 188 206 224 240</spanx></c> <c>25</c> <c><spanx style="vbare"> 8 15 34 50 67 83 99 115 131 146 162 178 193 209 224 239</spanx></c> <c>26</c> <c><spanx style="vbare">13 16 41 66 73 86 95 111 128 137 150 163 183 206 225 241</spanx></c> <c>27</c> <c><spanx style="vbare">17 25 37 52 63 75 92 102 119 132 144 160 175 191 212 231</spanx></c> <c>28</c> <c><spanx style="vbare">19 31 49 65 83 100 117 133 147 161 174 187 200 213 227 242</spanx></c> <c>29</c> <c><spanx style="vbare">18 31 52 68 88 103 117 126 138 149 163 177 192 207 223 239</spanx></c> <c>30</c> <c><spanx style="vbare">16 29 47 61 76 90 106 119 133 147 161 176 193 209 224 240</spanx></c> <c>31</c> <c><spanx style="vbare">15 21 35 50 61 73 86 97 110 119 129 141 175 198 218 237</spanx></c> </texttable> <t> Given the stage-1 codebook entry cb1_Q8[], the stage-2 residual res_Q10[], and their corresponding weights, w_Q9[], the reconstructed normalized LSF coefficients are <figure align="center"> <artwork align="center"><![CDATA[ NLSF_Q15[k] = (cb1_Q8[k]<<7) + (res_Q10[k]<<14)/w_Q9[k] , ]]></artwork> </figure> where the division is exact integer division. However, nothing in either the reconstruction process or the quantization process in the encoder thus far guarantees that the coefficients are monotonically increasing and separated well enough to ensure a stable filter. When using the reference encoder, roughly 2% of frames violate this constraint. The next section describes a stabilization procedure used to make these guarantees. </t> </section> <section anchor="silk_nlsf_stabilization" title="Normalized LSF Stabilization"> <!--TODO: Clean up lsf_stabilizer_overview_section--> <t> The normalized LSF stabilization procedure is implemented in silk_NLSF_stabilize() (NLSF_stabilize.c). This process ensures that consecutive values of the normalized LSF coefficients, NLSF_Q15[], are spaced some minimum distance apart (predetermined to be the 0.01 percentile of a large training set). <xref target="silk_nlsf_min_spacing"/> gives the minimum spacings for NB and MB and those for WB, where row k is the minimum allowed value of NLSF_Q[k]-NLSF_Q[k-1]. For the purposes of computing this spacing for the first and last coefficient, NLSF_Q15[-1] is taken to be 0, and NLSF_Q15[d_LPC] is taken to be 32768. </t> <texttable anchor="silk_nlsf_min_spacing" title="Minimum Spacing for Normalized LSF Coefficients"> <ttcol>Coefficient</ttcol> <ttcol align="right">NB and MB</ttcol> <ttcol align="right">WB</ttcol> <c>0</c> <c>250</c> <c>100</c> <c>1</c> <c>3</c> <c>3</c> <c>2</c> <c>6</c> <c>40</c> <c>3</c> <c>3</c> <c>3</c> <c>4</c> <c>3</c> <c>3</c> <c>5</c> <c>3</c> <c>3</c> <c>6</c> <c>4</c> <c>5</c> <c>7</c> <c>3</c> <c>14</c> <c>8</c> <c>3</c> <c>14</c> <c>9</c> <c>3</c> <c>10</c> <c>10</c> <c>461</c> <c>11</c> <c>11</c> <c/> <c>3</c> <c>12</c> <c/> <c>8</c> <c>13</c> <c/> <c>9</c> <c>14</c> <c/> <c>7</c> <c>15</c> <c/> <c>3</c> <c>16</c> <c/> <c>347</c> </texttable> <t> The procedure starts off by trying to make small adjustments which attempt to minimize the amount of distortion introduced. After 20 such adjustments, it falls back to a more direct method which guarantees the constraints are enforced but may require large adjustments. </t> <t> Let NDeltaMin_Q15[k] be the minimum required spacing for the current audio bandwidth from <xref target="silk_nlsf_min_spacing"/>. First, the procedure finds the index i where NLSF_Q15[i] - NLSF_Q15[i-1] - NDeltaMin_Q15[i] is the smallest, breaking ties by using the lower value of i. If this value is non-negative, then the stabilization stops; the coefficients satisfy all the constraints. Otherwise, if i == 0, it sets NLSF_Q15[0] to NDeltaMin_Q15[0], and if i == d_LPC, it sets NLSF_Q15[d_LPC-1] to (32768 - NDeltaMin_Q15[d_LPC]). For all other values of i, both NLSF_Q15[i-1] and NLSF_Q15[i] are updated as follows: <figure align="center"> <artwork align="center"><![CDATA[ i-1 __ min_center_Q15 = (NDeltaMin[i]>>1) + \ NDeltaMin[k] /_ k=0 d_LPC __ max_center_Q15 = 32768 - (NDeltaMin[i]>>1) - \ NDeltaMin[k] /_ k=i+1 center_freq_Q15 = clamp(min_center_Q15[i], (NLSF_Q15[i-1] + NLSF_Q15[i] + 1)>>1, max_center_Q15[i]) NLSF_Q15[i-1] = center_freq_Q15 - (NDeltaMin_Q15[i]>>1) NLSF_Q15[i] = NLSF_Q15[i-1] + NDeltaMin_Q15[i] . ]]></artwork> </figure> Then the procedure repeats again, until it has either executed 20 times or has stopped because the coefficients satisfy all the constraints. </t> <t> After the 20th repetition of the above procedure, the following fallback procedure executes once. First, the values of NLSF_Q15[k] for 0 <= k < d_LPC are sorted in ascending order. Then for each value of k from 0 to d_LPC-1, NLSF_Q15[k] is set to <figure align="center"> <artwork align="center"><![CDATA[ max(NLSF_Q15[k], NLSF_Q15[k-1] + NDeltaMin_Q15[k]) . ]]></artwork> </figure> Next, for each value of k from d_LPC-1 down to 0, NLSF_Q15[k] is set to <figure align="center"> <artwork align="center"><![CDATA[ min(NLSF_Q15[k], NLSF_Q15[k+1] - NDeltaMin_Q15[k+1]) . ]]></artwork> </figure> </t> </section> <section anchor="silk_nlsf_interpolation" title="Normalized LSF Interpolation"> <t> For 20 ms SILK frames, the first half of the frame (i.e., the first two subframes) may use normalized LSF coefficients that are interpolated between the decoded LSFs for the most recent coded frame (in the same channel) and the current frame. A Q2 interpolation factor follows the LSF coefficient indices in the bitstream, which is decoded using the PDF in <xref target="silk_nlsf_interp_pdf"/>. This happens in silk_decode_indices() (decode_indices.c). For the first frame after a decoder reset, when no prior LSF coefficients are available, the decoder still decodes this factor, but ignores its value and always uses 4 instead. For 10 ms SILK frames, this factor is not stored at all. </t> <texttable anchor="silk_nlsf_interp_pdf" title="PDF for Normalized LSF Interpolation Index"> <ttcol>PDF</ttcol> <c>{13, 22, 29, 11, 181}/256</c> </texttable> <t> Let n2_Q15[k] be the normalized LSF coefficients decoded by the procedure in <xref target="silk_nlsfs"/>, n0_Q15[k] be the LSF coefficients decoded for the prior frame, and w_Q2 be the interpolation factor. Then the normalized LSF coefficients used for the first half of a 20 ms frame, n1_Q15[k], are <figure align="center"> <artwork align="center"><![CDATA[ n1_Q15[k] = n0_Q15[k] + (w_Q2*(n2_Q15[k] - n0_Q15[k]) >> 2) . ]]></artwork> </figure> This interpolation is performed in silk_decode_parameters() (decode_parameters.c). </t> </section> <section anchor="silk_nlsf2lpc" title="Converting Normalized LSFs to LPC Coefficients"> <t> Any LPC filter A(z) can be split into a symmetric part P(z) and an anti-symmetric part Q(z) such that <figure align="center"> <artwork align="center"><![CDATA[ d_LPC __ -k 1 A(z) = 1 - \ a[k] * z = - * (P(z) + Q(z)) /_ 2 k=1 ]]></artwork> </figure> with <figure align="center"> <artwork align="center"><![CDATA[ -d_LPC-1 -1 P(z) = A(z) + z * A(z ) -d_LPC-1 -1 Q(z) = A(z) - z * A(z ) . ]]></artwork> </figure> The even normalized LSF coefficients correspond to a pair of conjugate roots of P(z), while the odd coefficients correspond to a pair of conjugate roots of Q(z), all of which lie on the unit circle. In addition, P(z) has a root at pi and Q(z) has a root at 0. Thus, they may be reconstructed mathematically from a set of normalized LSF coefficients, n[k], as <figure align="center"> <artwork align="center"><![CDATA[ d_LPC/2-1 -1 ___ -1 -2 P(z) = (1 + z ) * | | (1 - 2*cos(pi*n[2*k])*z + z ) k=0 d_LPC/2-1 -1 ___ -1 -2 Q(z) = (1 - z ) * | | (1 - 2*cos(pi*n[2*k+1])*z + z ) k=0 ]]></artwork> </figure> </t> <t> However, SILK performs this reconstruction using a fixed-point approximation so that all decoders can reproduce it in a bit-exact manner to avoid prediction drift. The function silk_NLSF2A() (NLSF2A.c) implements this procedure. </t> <t> To start, it approximates cos(pi*n[k]) using a table lookup with linear interpolation. The encoder SHOULD use the inverse of this piecewise linear approximation, rather than the true inverse of the cosine function, when deriving the normalized LSF coefficients. </t> <t> The top 7 bits of each normalized LSF coefficient index a value in the table, and the next 8 bits interpolate between it and the next value. Let i = n[k]>>8 be the integer index and f = n[k]&255 be the fractional part of a given coefficient. Then the approximated cosine, c_Q17[k], is <figure align="center"> <artwork align="center"><![CDATA[ c_Q17[k] = (cos_Q13[i]*256 + (cos_Q13[i+1]-cos_Q13[i])*f + 8) >> 4 , ]]></artwork> </figure> where cos_Q13[i] is the corresponding entry of <xref target="silk_cos_table"/>. </t> <texttable anchor="silk_cos_table" title="Q13 Cosine Table for LSF Conversion"> <ttcol align="right"></ttcol> <ttcol align="right">0</ttcol> <ttcol align="right">1</ttcol> <ttcol align="right">2</ttcol> <ttcol align="right">3</ttcol> <c>0</c> <c>8192</c> <c>8190</c> <c>8182</c> <c>8170</c> <c>4</c> <c>8152</c> <c>8130</c> <c>8104</c> <c>8072</c> <c>8</c> <c>8034</c> <c>7994</c> <c>7946</c> <c>7896</c> <c>12</c> <c>7840</c> <c>7778</c> <c>7714</c> <c>7644</c> <c>16</c> <c>7568</c> <c>7490</c> <c>7406</c> <c>7318</c> <c>20</c> <c>7226</c> <c>7128</c> <c>7026</c> <c>6922</c> <c>24</c> <c>6812</c> <c>6698</c> <c>6580</c> <c>6458</c> <c>28</c> <c>6332</c> <c>6204</c> <c>6070</c> <c>5934</c> <c>32</c> <c>5792</c> <c>5648</c> <c>5502</c> <c>5352</c> <c>36</c> <c>5198</c> <c>5040</c> <c>4880</c> <c>4718</c> <c>40</c> <c>4552</c> <c>4382</c> <c>4212</c> <c>4038</c> <c>44</c> <c>3862</c> <c>3684</c> <c>3502</c> <c>3320</c> <c>48</c> <c>3136</c> <c>2948</c> <c>2760</c> <c>2570</c> <c>52</c> <c>2378</c> <c>2186</c> <c>1990</c> <c>1794</c> <c>56</c> <c>1598</c> <c>1400</c> <c>1202</c> <c>1002</c> <c>60</c> <c>802</c> <c>602</c> <c>402</c> <c>202</c> <c>64</c> <c>0</c> <c>-202</c> <c>-402</c> <c>-602</c> <c>68</c> <c>-802</c><c>-1002</c><c>-1202</c><c>-1400</c> <c>72</c> <c>-1598</c><c>-1794</c><c>-1990</c><c>-2186</c> <c>76</c> <c>-2378</c><c>-2570</c><c>-2760</c><c>-2948</c> <c>80</c> <c>-3136</c><c>-3320</c><c>-3502</c><c>-3684</c> <c>84</c> <c>-3862</c><c>-4038</c><c>-4212</c><c>-4382</c> <c>88</c> <c>-4552</c><c>-4718</c><c>-4880</c><c>-5040</c> <c>92</c> <c>-5198</c><c>-5352</c><c>-5502</c><c>-5648</c> <c>96</c> <c>-5792</c><c>-5934</c><c>-6070</c><c>-6204</c> <c>100</c> <c>-6332</c><c>-6458</c><c>-6580</c><c>-6698</c> <c>104</c> <c>-6812</c><c>-6922</c><c>-7026</c><c>-7128</c> <c>108</c> <c>-7226</c><c>-7318</c><c>-7406</c><c>-7490</c> <c>112</c> <c>-7568</c><c>-7644</c><c>-7714</c><c>-7778</c> <c>116</c> <c>-7840</c><c>-7896</c><c>-7946</c><c>-7994</c> <c>120</c> <c>-8034</c><c>-8072</c><c>-8104</c><c>-8130</c> <c>124</c> <c>-8152</c><c>-8170</c><c>-8182</c><c>-8190</c> <c>128</c> <c>-8192</c> <c/> <c/> <c/> </texttable> <t> Given the list of cosine values, silk_NLSF2A_find_poly() (NLSF2A.c) computes the coefficients of P and Q, described here via a simple recurrence. Let p_Q16[k][j] and q_Q16[k][j] be the coefficients of the products of the first (k+1) root pairs for P and Q, with j indexing the coefficient number. Only the first (k+2) coefficients are needed, as the products are symmetric. Let p_Q16[0][0] = q_Q16[0][0] = 1<<16, p_Q16[0][1] = -c_Q17[0], q_Q16[0][1] = -c_Q17[1], and d2 = d_LPC/2. As boundary conditions, assume p_Q16[k][j] = q_Q16[k][j] = 0 for all j < 0. Also, assume p_Q16[k][k+2] = p_Q16[k][k] and q_Q16[k][k+2] = q_Q16[k][k] (because of the symmetry). Then, for 0 <k < d2 and 0 <= j <= k+1, <figure align="center"> <artwork align="center"><![CDATA[ p_Q16[k][j] = p_Q16[k-1][j] + p_Q16[k-1][j-2] - ((c_Q17[2*k]*p_Q16[k-1][j-1] + 32768)>>16) , q_Q16[k][j] = q_Q16[k-1][j] + q_Q16[k-1][j-2] - ((c_Q17[2*k+1]*q_Q16[k-1][j-1] + 32768)>>16) . ]]></artwork> </figure> The use of Q17 values for the cosine terms in an otherwise Q16 expression implicitly scales them by a factor of 2. The multiplications in this recurrence may require up to 48 bits of precision in the result to avoid overflow. In practice, each row of the recurrence only depends on the previous row, so an implementation does not need to store all of them. </t> <t> silk_NLSF2A() uses the values from the last row of this recurrence to reconstruct a 32-bit version of the LPC filter (without the leading 1.0 coefficient), a32_Q17[k], 0 <= k < d2: <figure align="center"> <artwork align="center"><![CDATA[ a32_Q17[k] = -(q_Q16[d2-1][k+1] - q_Q16[d2-1][k]) - (p_Q16[d2-1][k+1] + p_Q16[d2-1][k])) , a32_Q17[d_LPC-k-1] = (q_Q16[d2-1][k+1] - q_Q16[d2-1][k]) - (p_Q16[d2-1][k+1] + p_Q16[d2-1][k])) . ]]></artwork> </figure> The sum and difference of two terms from each of the p_Q16 and q_Q16 coefficient lists reflect the (1 + z**-1) and (1 - z**-1) factors of P and Q, respectively. The promotion of the expression from Q16 to Q17 implicitly scales the result by 1/2. </t> </section> <section anchor="silk_lpc_range_limit" title="Limiting the Range of the LPC Coefficients"> <t> The a32_Q17[] coefficients are too large to fit in a 16-bit value, which significantly increases the cost of applying this filter in fixed-point decoders. Reducing them to Q12 precision doesn't incur any significant quality loss, but still does not guarantee they will fit. silk_NLSF2A() applies up to 10 rounds of bandwidth expansion to limit the dynamic range of these coefficients. Even floating-point decoders SHOULD perform these steps, to avoid mismatch. </t> <t> For each round, the process first finds the index k such that abs(a32_Q17[k]) is largest, breaking ties by choosing the lowest value of k. Then, it computes the corresponding Q12 precision value, maxabs_Q12, subject to an upper bound to avoid overflow in subsequent computations: <figure align="center"> <artwork align="center"><![CDATA[ maxabs_Q12 = min((maxabs_Q17 + 16) >> 5, 163838) . ]]></artwork> </figure> If this is larger than 32767, the procedure derives the chirp factor, sc_Q16[0], to use in the bandwidth expansion as <figure align="center"> <artwork align="center"><![CDATA[ (maxabs_Q12 - 32767) << 14 sc_Q16[0] = 65470 - -------------------------- , (maxabs_Q12 * (k+1)) >> 2 ]]></artwork> </figure> where the division here is exact integer division. This is an approximation of the chirp factor needed to reduce the target coefficient to 32767, though it is both less than 0.999 and, for k > 0 when maxabs_Q12 is much greater than 32767, still slightly too large. </t> <t> silk_bwexpander_32() (bwexpander_32.c) performs the bandwidth expansion (again, only when maxabs_Q12 is greater than 32767) using the following recurrence: <figure align="center"> <artwork align="center"><![CDATA[ a32_Q17[k] = (a32_Q17[k]*sc_Q16[k]) >> 16 sc_Q16[k+1] = (sc_Q16[0]*sc_Q16[k] + 32768) >> 16 ]]></artwork> </figure> The first multiply may require up to 48 bits of precision in the result to avoid overflow. The second multiply must be unsigned to avoid overflow with only 32 bits of precision. The reference implementation uses a slightly more complex formulation that avoids the 32-bit overflow using signed multiplication, but is otherwise equivalent. </t> <t> After 10 rounds of bandwidth expansion are performed, they are simply saturated to 16 bits: <figure align="center"> <artwork align="center"><![CDATA[ a32_Q17[k] = clamp(-32768, (a32_Q17[k]+16) >> 5, 32767) << 5 . ]]></artwork> </figure> Because this performs the actual saturation in the Q12 domain, but converts the coefficients back to the Q17 domain for the purposes of prediction gain limiting, this step must be performed after the 10th round of bandwidth expansion, regardless of whether or not the Q12 version of any coefficient still overflows a 16-bit integer. This saturation is not performed if maxabs_Q12 drops to 32767 or less prior to the 10th round. </t> </section> <section anchor="silk_lpc_gain_limit" title="Limiting the Prediction Gain of the LPC Filter"> <t> The prediction gain of an LPC synthesis filter is the square-root of the output energy when the filter is excited by a unit-energy impulse. Even if the Q12 coefficients would fit, the resulting filter may still have a significant gain (especially for voiced sounds), making the filter unstable. silk_NLSF2A() applies up to 18 additional rounds of bandwidth expansion to limit the prediction gain. Instead of controlling the amount of bandwidth expansion using the prediction gain itself (which may diverge to infinity for an unstable filter), silk_NLSF2A() uses LPC_inverse_pred_gain_QA() (LPC_inv_pred_gain.c) to compute the reflection coefficients associated with the filter. The filter is stable if and only if the magnitude of these coefficients is sufficiently less than one. The reflection coefficients, rc[k], can be computed using a simple Levinson recurrence, initialized with the LPC coefficients a[d_LPC-1][n] = a[n], and then updated via <figure align="center"> <artwork align="center"><![CDATA[ rc[k] = -a[k][k] , a[k][n] - a[k][k-n-1]*rc[k] a[k-1][n] = --------------------------- . 2 1 - rc[k] ]]></artwork> </figure> </t> <t> However, LPC_inverse_pred_gain_QA() approximates this using fixed-point arithmetic to guarantee reproducible results across platforms and implementations. It is important to run on the real Q12 coefficients that will be used during reconstruction, because small changes in the coefficients can make a stable filter unstable, but increasing the precision back to Q16 allows more accurate computation of the reflection coefficients. Thus, let <figure align="center"> <artwork align="center"><![CDATA[ a32_Q16[d_LPC-1][n] = ((a32_Q17[n] + 16) >> 5) << 4 ]]></artwork> </figure> be the Q16 representation of the Q12 version of the LPC coefficients that will eventually be used. Then for each k from d_LPC-1 down to 0, if abs(a32_Q16[k][k]) > 65520, the filter is unstable and the recurrence stops. Otherwise, the row k-1 of a32_Q16 is computed from row k as <figure align="center"> <artwork align="center"><![CDATA[ rc_Q31[k] = -a32_Q16[k][k] << 15 , div_Q30[k] = (1<<30) - 1 - (rc_Q31[k]*rc_Q31[k] >> 32) , b1[k] = ilog(div_Q30[k]) - 16 , (1<<29) - 1 inv_Qb1[k] = ----------------------- , div_Q30[k] >> (b1[k]+1) err_Q29[k] = (1<<29) - ((div_Q30[k]<<(15-b1[k]))*inv_Qb1[k] >> 16) , mul_Q16[k] = ((inv_Qb1[k] << 16) + (err_Q29[k]*inv_Qb1[k] >> 13)) >> b1[k] , b2[k] = ilog(mul_Q16[k]) - 15 , t_Q16[k-1][n] = a32_Q16[k][n] - ((a32_Q16[k][k-n-1]*rc_Q31[k] >> 32) << 1) , a32_Q16[k-1][n] = ((t_Q16[k-1][n] * (mul_Q16[k] << (16-b2[k]))) >> 32) << b2[k] . ]]></artwork> </figure> Here, rc_Q30[k] are the reflection coefficients. div_Q30[k] is the denominator for each iteration, and mul_Q16[k] is its multiplicative inverse. inv_Qb1[k], which ranges from 16384 to 32767, is a low-precision version of that inverse (with b1[k] fractional bits, where b1[k] ranges from 3 to 14). err_Q29[k] is the residual error, ranging from -32392 to 32763, which is used to improve the accuracy. t_Q16[k-1][n], 0 <= n < k, are the numerators for the next row of coefficients in the recursion, and a32_Q16[k-1][n] is the final version of that row. Every multiply in this procedure except the one used to compute mul_Q16[k] requires more than 32 bits of precision, but otherwise all intermediate results fit in 32 bits or less. In practice, because each row only depends on the next one, an implementation does not need to store them all. If abs(a32_Q16[k][k]) <= 65520 for 0 <= k < d_LPC, then the filter is considered stable. </t> <t> On round i, 1 <= i <= 18, if the filter passes this stability check, then this procedure stops, and the final LPC coefficients to use for reconstruction in <xref target="silk_lpc_synthesis"/> are <figure align="center"> <artwork align="center"><![CDATA[ a_Q12[k] = (a32_Q17[k] + 16) >> 5 . ]]></artwork> </figure> Otherwise, a round of bandwidth expansion is applied using the same procedure as in <xref target="silk_lpc_range_limit"/>, with <figure align="center"> <artwork align="center"><![CDATA[ sc_Q16[0] = 65536 - i*(i+9) . ]]></artwork> </figure> If, after the 18th round, the filter still fails the stability check, then a_Q12[k] is set to 0 for all k. </t> </section> </section> <section anchor="silk_ltp_params" title="Long-Term Prediction (LTP) Parameters"> <t> After the normalized LSF indices and, for 20 ms frames, the LSF interpolation index, voiced frames (see <xref target="silk_frame_type"/>) include additional Long-Term Prediction (LTP) parameters. There is one primary lag index for each SILK frame, but this is refined to produce a separate lag index per subframe using a vector quantizer. Each subframe also gets its own prediction gain coefficient. </t> <section anchor="silk_ltp_lags" title="Pitch Lags"> <t> The primary lag index is coded either relative to the primary lag of the prior frame or as an absolute index. Like the quantization gains, the primary pitch lag is coded either as an absolute index, or relative to the most recent coded frame in the same channel. Absolute coding is used if and only if <list style="symbols"> <t>This is the first LBRR frame for this channel in the current Opus frame,</t> <t> This is an LBRR frame where the LBRR flags (see <xref target="silk_header_bits"/> and <xref target="silk_lbrr_flags"/>) indicate the previous LBRR frame in the same channel is not coded, </t> <t> This is the first regular SILK frame for this channel in the current Opus frame, or </t> <t> The most recently coded frame in the current channel was not voiced (see <xref target="silk_frame_type"/>). </t> </list> In particular, unlike an LBRR frame where the previous frame is not coded, in a 60 ms stereo Opus frame, if the first and third regular SILK frames in the side channel are coded, voiced frames, but the second is not coded, the third still uses relative coding. </t> <t> With absolute coding, the primary pitch lag may range from 2 ms (inclusive) up to 18 ms (exclusive), corresponding to pitches from 500 Hz down to 55.6 Hz, respectively. It is comprised of a high part and a low part, where the decoder reads the high part using the 32-entry codebook in <xref target="silk_abs_pitch_high_pdf"/> and the low part using the codebook corresponding to the current audio bandwidth from <xref target="silk_abs_pitch_low_pdf"/>. The final primary pitch lag is then <figure align="center"> <artwork align="center"><![CDATA[ lag = lag_high*lag_scale + lag_low + lag_min ]]></artwork> </figure> where lag_high is the high part, lag_low is the low part, and lag_scale and lag_min are the values from the "Scale" and "Minimum Lag" columns of <xref target="silk_abs_pitch_low_pdf"/>, respectively. </t> <texttable anchor="silk_abs_pitch_high_pdf" title="PDF for High Part of Primary Pitch Lag"> <ttcol align="left">PDF</ttcol> <c>{3, 3, 6, 11, 21, 30, 32, 19, 11, 10, 12, 13, 13, 12, 11, 9, 8, 7, 6, 4, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1}/256</c> </texttable> <texttable anchor="silk_abs_pitch_low_pdf" title="PDF for Low Part of Primary Pitch Lag"> <ttcol>Audio Bandwidth</ttcol> <ttcol>PDF</ttcol> <ttcol>Scale</ttcol> <ttcol>Minimum Lag</ttcol> <ttcol>Maximum Lag</ttcol> <c>NB</c> <c>{64, 64, 64, 64}/256</c> <c>4</c> <c>16</c> <c>144</c> <c>MB</c> <c>{43, 42, 43, 43, 42, 43}/256</c> <c>6</c> <c>24</c> <c>216</c> <c>WB</c> <c>{32, 32, 32, 32, 32, 32, 32, 32}/256</c> <c>8</c> <c>32</c> <c>288</c> </texttable> <t> All frames that do not use absolute coding for the primary lag index use relative coding instead. The decoder reads a single delta value using the 21-entry PDF in <xref target="silk_rel_pitch_pdf"/>. If the resulting value is zero, it falls back to the absolute coding procedure from the prior paragraph. Otherwise, the final primary pitch lag is then <figure align="center"> <artwork align="center"><![CDATA[ lag = lag_prev + (delta_lag_index - 9) ]]></artwork> </figure> where lag_prev is the primary pitch lag from the most recent frame in the same channel and delta_lag_index is the value just decoded. This allows a per-frame change in the pitch lag of -8 to +11 samples. The decoder does no clamping at this point, so this value can fall outside the range of 2 ms to 18 ms, and the decoder must use this unclamped value when using relative coding in the next SILK frame (if any). However, because an Opus frame can use relative coding for at most two consecutive SILK frames, integer overflow should not be an issue. </t> <texttable anchor="silk_rel_pitch_pdf" title="PDF for Primary Pitch Lag Change"> <ttcol align="left">PDF</ttcol> <c>{46, 2, 2, 3, 4, 6, 10, 15, 26, 38, 30, 22, 15, 10, 7, 6, 4, 4, 2, 2, 2}/256</c> </texttable> <t> After the primary pitch lag, a "pitch contour", stored as a single entry from one of four small VQ codebooks, gives lag offsets for each subframe in the current SILK frame. The codebook index is decoded using one of the PDFs in <xref target="silk_pitch_contour_pdfs"/> depending on the current frame size and audio bandwidth. Tables <xref format="counter" target="silk_pitch_contour_cb_nb10ms"/> through <xref format="counter" target="silk_pitch_contour_cb_mbwb20ms"/> give the corresponding offsets to apply to the primary pitch lag for each subframe given the decoded codebook index. </t> <texttable anchor="silk_pitch_contour_pdfs" title="PDFs for Subframe Pitch Contour"> <ttcol>Audio Bandwidth</ttcol> <ttcol>SILK Frame Size</ttcol> <ttcol align="right">Codebook Size</ttcol> <ttcol>PDF</ttcol> <c>NB</c> <c>10 ms</c> <c>3</c> <c>{143, 50, 63}/256</c> <c>NB</c> <c>20 ms</c> <c>11</c> <c>{68, 12, 21, 17, 19, 22, 30, 24, 17, 16, 10}/256</c> <c>MB or WB</c> <c>10 ms</c> <c>12</c> <c>{91, 46, 39, 19, 14, 12, 8, 7, 6, 5, 5, 4}/256</c> <c>MB or WB</c> <c>20 ms</c> <c>34</c> <c>{33, 22, 18, 16, 15, 14, 14, 13, 13, 10, 9, 9, 8, 6, 6, 6, 5, 4, 4, 4, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1}/256</c> </texttable> <texttable anchor="silk_pitch_contour_cb_nb10ms" title="Codebook Vectors for Subframe Pitch Contour: NB, 10 ms Frames"> <ttcol>Index</ttcol> <ttcol align="right">Subframe Offsets</ttcol> <c>0</c> <c><spanx style="vbare"> 0 0</spanx></c> <c>1</c> <c><spanx style="vbare"> 1 0</spanx></c> <c>2</c> <c><spanx style="vbare"> 0 1</spanx></c> </texttable> <texttable anchor="silk_pitch_contour_cb_nb20ms" title="Codebook Vectors for Subframe Pitch Contour: NB, 20 ms Frames"> <ttcol>Index</ttcol> <ttcol align="right">Subframe Offsets</ttcol> <c>0</c> <c><spanx style="vbare"> 0 0 0 0</spanx></c> <c>1</c> <c><spanx style="vbare"> 2 1 0 -1</spanx></c> <c>2</c> <c><spanx style="vbare">-1 0 1 2</spanx></c> <c>3</c> <c><spanx style="vbare">-1 0 0 1</spanx></c> <c>4</c> <c><spanx style="vbare">-1 0 0 0</spanx></c> <c>5</c> <c><spanx style="vbare"> 0 0 0 1</spanx></c> <c>6</c> <c><spanx style="vbare"> 0 0 1 1</spanx></c> <c>7</c> <c><spanx style="vbare"> 1 1 0 0</spanx></c> <c>8</c> <c><spanx style="vbare"> 1 0 0 0</spanx></c> <c>9</c> <c><spanx style="vbare"> 0 0 0 -1</spanx></c> <c>10</c> <c><spanx style="vbare"> 1 0 0 -1</spanx></c> </texttable> <texttable anchor="silk_pitch_contour_cb_mbwb10ms" title="Codebook Vectors for Subframe Pitch Contour: MB or WB, 10 ms Frames"> <ttcol>Index</ttcol> <ttcol align="right">Subframe Offsets</ttcol> <c>0</c> <c><spanx style="vbare"> 0 0</spanx></c> <c>1</c> <c><spanx style="vbare"> 0 1</spanx></c> <c>2</c> <c><spanx style="vbare"> 1 0</spanx></c> <c>3</c> <c><spanx style="vbare">-1 1</spanx></c> <c>4</c> <c><spanx style="vbare"> 1 -1</spanx></c> <c>5</c> <c><spanx style="vbare">-1 2</spanx></c> <c>6</c> <c><spanx style="vbare"> 2 -1</spanx></c> <c>7</c> <c><spanx style="vbare">-2 2</spanx></c> <c>8</c> <c><spanx style="vbare"> 2 -2</spanx></c> <c>9</c> <c><spanx style="vbare">-2 3</spanx></c> <c>10</c> <c><spanx style="vbare"> 3 -2</spanx></c> <c>11</c> <c><spanx style="vbare">-3 3</spanx></c> </texttable> <texttable anchor="silk_pitch_contour_cb_mbwb20ms" title="Codebook Vectors for Subframe Pitch Contour: MB or WB, 20 ms Frames"> <ttcol>Index</ttcol> <ttcol align="right">Subframe Offsets</ttcol> <c>0</c> <c><spanx style="vbare"> 0 0 0 0</spanx></c> <c>1</c> <c><spanx style="vbare"> 0 0 1 1</spanx></c> <c>2</c> <c><spanx style="vbare"> 1 1 0 0</spanx></c> <c>3</c> <c><spanx style="vbare">-1 0 0 0</spanx></c> <c>4</c> <c><spanx style="vbare"> 0 0 0 1</spanx></c> <c>5</c> <c><spanx style="vbare"> 1 0 0 0</spanx></c> <c>6</c> <c><spanx style="vbare">-1 0 0 1</spanx></c> <c>7</c> <c><spanx style="vbare"> 0 0 0 -1</spanx></c> <c>8</c> <c><spanx style="vbare">-1 0 1 2</spanx></c> <c>9</c> <c><spanx style="vbare"> 1 0 0 -1</spanx></c> <c>10</c> <c><spanx style="vbare">-2 -1 1 2</spanx></c> <c>11</c> <c><spanx style="vbare"> 2 1 0 -1</spanx></c> <c>12</c> <c><spanx style="vbare">-2 0 0 2</spanx></c> <c>13</c> <c><spanx style="vbare">-2 0 1 3</spanx></c> <c>14</c> <c><spanx style="vbare"> 2 1 -1 -2</spanx></c> <c>15</c> <c><spanx style="vbare">-3 -1 1 3</spanx></c> <c>16</c> <c><spanx style="vbare"> 2 0 0 -2</spanx></c> <c>17</c> <c><spanx style="vbare"> 3 1 0 -2</spanx></c> <c>18</c> <c><spanx style="vbare">-3 -1 2 4</spanx></c> <c>19</c> <c><spanx style="vbare">-4 -1 1 4</spanx></c> <c>20</c> <c><spanx style="vbare"> 3 1 -1 -3</spanx></c> <c>21</c> <c><spanx style="vbare">-4 -1 2 5</spanx></c> <c>22</c> <c><spanx style="vbare"> 4 2 -1 -3</spanx></c> <c>23</c> <c><spanx style="vbare"> 4 1 -1 -4</spanx></c> <c>24</c> <c><spanx style="vbare">-5 -1 2 6</spanx></c> <c>25</c> <c><spanx style="vbare"> 5 2 -1 -4</spanx></c> <c>26</c> <c><spanx style="vbare">-6 -2 2 6</spanx></c> <c>27</c> <c><spanx style="vbare">-5 -2 2 5</spanx></c> <c>28</c> <c><spanx style="vbare"> 6 2 -1 -5</spanx></c> <c>29</c> <c><spanx style="vbare">-7 -2 3 8</spanx></c> <c>30</c> <c><spanx style="vbare"> 6 2 -2 -6</spanx></c> <c>31</c> <c><spanx style="vbare"> 5 2 -2 -5</spanx></c> <c>32</c> <c><spanx style="vbare"> 8 3 -2 -7</spanx></c> <c>33</c> <c><spanx style="vbare">-9 -3 3 9</spanx></c> </texttable> <t> The final pitch lag for each subframe is assembled in silk_decode_pitch() (decode_pitch.c). Let lag be the primary pitch lag for the current SILK frame, contour_index be index of the VQ codebook, and lag_cb[contour_index][k] be the corresponding entry of the codebook from the appropriate table given above for the k'th subframe. Then the final pitch lag for that subframe is <figure align="center"> <artwork align="center"><![CDATA[ pitch_lags[k] = clamp(lag_min, lag + lag_cb[contour_index][k], lag_max) ]]></artwork> </figure> where lag_min and lag_max are the values from the "Minimum Lag" and "Maximum Lag" columns of <xref target="silk_abs_pitch_low_pdf"/>, respectively. </t> </section> <section anchor="silk_ltp_coeffs" title="LTP Filter Coefficients"> <t> SILK can use a separate 5-tap pitch filter for each subframe. It selects the filter to use from one of three codebooks. The three codebooks each represent different rate-distortion trade-offs, with average rates of 1.61 bits/subframe, 3.68 bits/subframe, and 4.85 bits/subframe, respectively. </t> <t> The importance of the filter coefficients generally depends on two factors: the periodicity of the signal and relative energy between the current subframe and the signal from one period earlier. Greater periodicity and decaying energy both lead to more important filter coefficients, and thus should be coded with lower distortion and higher rate. These properties are relatively stable over the duration of a single SILK frame, hence all of the subframes in a SILK frame must choose their filter from the same codebook. This is signaled with an explicitly-coded "periodicity index". This immediately follows the subframe pitch lags, and is coded using the 3-entry PDF from <xref target="silk_perindex_pdf"/>. </t> <texttable anchor="silk_perindex_pdf" title="Periodicity Index PDF"> <ttcol>PDF</ttcol> <c>{77, 80, 99}/256</c> </texttable> <t> The index of the filter to use for each subframe follows. They are all coded using the PDF from <xref target="silk_ltp_filter_pdfs"/> corresponding to the periodicity index. Tables <xref format="counter" target="silk_ltp_filter_coeffs0"/> through <xref format="counter" target="silk_ltp_filter_coeffs2"/> contain the corresponding filter taps as signed Q7 integers. </t> <texttable anchor="silk_ltp_filter_pdfs" title="LTP Filter PDFs"> <ttcol>Periodicity Index</ttcol> <ttcol align="right">Codebook Size</ttcol> <ttcol>PDF</ttcol> <c>0</c> <c>8</c> <c>{185, 15, 13, 13, 9, 9, 6, 6}/256</c> <c>1</c> <c>16</c> <c>{57, 34, 21, 20, 15, 13, 12, 13, 10, 10, 9, 10, 9, 8, 7, 8}/256</c> <c>2</c> <c>32</c> <c>{15, 16, 14, 12, 12, 12, 11, 11, 11, 10, 9, 9, 9, 9, 8, 8, 8, 8, 7, 7, 6, 6, 5, 4, 5, 4, 4, 4, 3, 4, 3, 2}/256</c> </texttable> <texttable anchor="silk_ltp_filter_coeffs0" title="Codebook Vectors for LTP Filter, Periodicity Index 0"> <ttcol>Index</ttcol> <ttcol align="right">Filter Taps (Q7)</ttcol> <c>0</c> <c><spanx style="vbare"> 4 6 24 7 5</spanx></c> <c>1</c> <c><spanx style="vbare"> 0 0 2 0 0</spanx></c> <c>2</c> <c><spanx style="vbare"> 12 28 41 13 -4</spanx></c> <c>3</c> <c><spanx style="vbare"> -9 15 42 25 14</spanx></c> <c>4</c> <c><spanx style="vbare"> 1 -2 62 41 -9</spanx></c> <c>5</c> <c><spanx style="vbare">-10 37 65 -4 3</spanx></c> <c>6</c> <c><spanx style="vbare"> -6 4 66 7 -8</spanx></c> <c>7</c> <c><spanx style="vbare"> 16 14 38 -3 33</spanx></c> </texttable> <texttable anchor="silk_ltp_filter_coeffs1" title="Codebook Vectors for LTP Filter, Periodicity Index 1"> <ttcol>Index</ttcol> <ttcol align="right">Filter Taps (Q7)</ttcol> <c>0</c> <c><spanx style="vbare"> 13 22 39 23 12</spanx></c> <c>1</c> <c><spanx style="vbare"> -1 36 64 27 -6</spanx></c> <c>2</c> <c><spanx style="vbare"> -7 10 55 43 17</spanx></c> <c>3</c> <c><spanx style="vbare"> 1 1 8 1 1</spanx></c> <c>4</c> <c><spanx style="vbare"> 6 -11 74 53 -9</spanx></c> <c>5</c> <c><spanx style="vbare">-12 55 76 -12 8</spanx></c> <c>6</c> <c><spanx style="vbare"> -3 3 93 27 -4</spanx></c> <c>7</c> <c><spanx style="vbare"> 26 39 59 3 -8</spanx></c> <c>8</c> <c><spanx style="vbare"> 2 0 77 11 9</spanx></c> <c>9</c> <c><spanx style="vbare"> -8 22 44 -6 7</spanx></c> <c>10</c> <c><spanx style="vbare"> 40 9 26 3 9</spanx></c> <c>11</c> <c><spanx style="vbare"> -7 20 101 -7 4</spanx></c> <c>12</c> <c><spanx style="vbare"> 3 -8 42 26 0</spanx></c> <c>13</c> <c><spanx style="vbare">-15 33 68 2 23</spanx></c> <c>14</c> <c><spanx style="vbare"> -2 55 46 -2 15</spanx></c> <c>15</c> <c><spanx style="vbare"> 3 -1 21 16 41</spanx></c> </texttable> <texttable anchor="silk_ltp_filter_coeffs2" title="Codebook Vectors for LTP Filter, Periodicity Index 2"> <ttcol>Index</ttcol> <ttcol align="right">Filter Taps (Q7)</ttcol> <c>0</c> <c><spanx style="vbare"> -6 27 61 39 5</spanx></c> <c>1</c> <c><spanx style="vbare">-11 42 88 4 1</spanx></c> <c>2</c> <c><spanx style="vbare"> -2 60 65 6 -4</spanx></c> <c>3</c> <c><spanx style="vbare"> -1 -5 73 56 1</spanx></c> <c>4</c> <c><spanx style="vbare"> -9 19 94 29 -9</spanx></c> <c>5</c> <c><spanx style="vbare"> 0 12 99 6 4</spanx></c> <c>6</c> <c><spanx style="vbare"> 8 -19 102 46 -13</spanx></c> <c>7</c> <c><spanx style="vbare"> 3 2 13 3 2</spanx></c> <c>8</c> <c><spanx style="vbare"> 9 -21 84 72 -18</spanx></c> <c>9</c> <c><spanx style="vbare">-11 46 104 -22 8</spanx></c> <c>10</c> <c><spanx style="vbare"> 18 38 48 23 0</spanx></c> <c>11</c> <c><spanx style="vbare">-16 70 83 -21 11</spanx></c> <c>12</c> <c><spanx style="vbare"> 5 -11 117 22 -8</spanx></c> <c>13</c> <c><spanx style="vbare"> -6 23 117 -12 3</spanx></c> <c>14</c> <c><spanx style="vbare"> 3 -8 95 28 4</spanx></c> <c>15</c> <c><spanx style="vbare">-10 15 77 60 -15</spanx></c> <c>16</c> <c><spanx style="vbare"> -1 4 124 2 -4</spanx></c> <c>17</c> <c><spanx style="vbare"> 3 38 84 24 -25</spanx></c> <c>18</c> <c><spanx style="vbare"> 2 13 42 13 31</spanx></c> <c>19</c> <c><spanx style="vbare"> 21 -4 56 46 -1</spanx></c> <c>20</c> <c><spanx style="vbare"> -1 35 79 -13 19</spanx></c> <c>21</c> <c><spanx style="vbare"> -7 65 88 -9 -14</spanx></c> <c>22</c> <c><spanx style="vbare"> 20 4 81 49 -29</spanx></c> <c>23</c> <c><spanx style="vbare"> 20 0 75 3 -17</spanx></c> <c>24</c> <c><spanx style="vbare"> 5 -9 44 92 -8</spanx></c> <c>25</c> <c><spanx style="vbare"> 1 -3 22 69 31</spanx></c> <c>26</c> <c><spanx style="vbare"> -6 95 41 -12 5</spanx></c> <c>27</c> <c><spanx style="vbare"> 39 67 16 -4 1</spanx></c> <c>28</c> <c><spanx style="vbare"> 0 -6 120 55 -36</spanx></c> <c>29</c> <c><spanx style="vbare">-13 44 122 4 -24</spanx></c> <c>30</c> <c><spanx style="vbare"> 81 5 11 3 7</spanx></c> <c>31</c> <c><spanx style="vbare"> 2 0 9 10 88</spanx></c> </texttable> </section> <section anchor="silk_ltp_scaling" title="LTP Scaling Parameter"> <t> An LTP scaling parameter appears after the LTP filter coefficients if and only if <list style="symbols"> <t>This is a voiced frame (see <xref target="silk_frame_type"/>), and</t> <t>Either <list style="symbols"> <t>This is the first LBRR frame for this channel in the current Opus frame,</t> <t> This is an LBRR frame where the LBRR flags (see <xref target="silk_header_bits"/> and <xref target="silk_lbrr_flags"/>) indicate the previous LBRR frame in the same channel is not coded, or </t> <t> This is the first regular SILK frame for this channel in the current Opus frame. </t> </list> </t> </list> This allows the encoder to trade off the prediction gain between packets against the recovery time after packet loss. Unlike absolute-coding for pitch lags, a regular SILK frame other than the first one in a channel will not include this field even if the prior frame was not voiced. </t> <t> If present, the value is coded using the 3-entry PDF in <xref target="silk_ltp_scaling_pdf"/>. The three possible values represent Q14 scale factors of 15565, 12288, and 8192, respectively (corresponding to approximately 0.95, 0.75, and 0.5). Frames that do not code the scaling parameter use the default factor of 15565 (approximately 0.95). </t> <texttable anchor="silk_ltp_scaling_pdf" title="PDF for LTP Scaling Parameter"> <ttcol align="left">PDF</ttcol> <c>{128, 64, 64}/256</c> </texttable> </section> </section> <section anchor="silk_seed" title="Linear Congruential Generator (LCG) Seed"> <t> SILK uses a linear congruential generator (LCG) to inject pseudorandom noise into the quantized excitation. To ensure synchronization of this process between the encoder and decoder, each SILK frame stores a 2-bit seed after the LTP parameters (if any). The encoder may consider the choice of this seed during quantization, meaning the flexibility to choose the LCG seed can reduce distortion. The seed is decoded with the uniform 4-entry PDF in <xref target="silk_seed_pdf"/>, yielding a value between 0 and 3, inclusive. </t> <texttable anchor="silk_seed_pdf" title="PDF for LCG Seed"> <ttcol align="left">PDF</ttcol> <c>{64, 64, 64, 64}/256</c> </texttable> </section> <section anchor="silk_excitation" title="Excitation"> <t> SILK codes the excitation using a modified version of the Pyramid Vector Quantization (PVQ) codebook <xref target="PVQ"/>. The PVQ codebook is designed for Laplace-distributed values and consists of all sums of K signed, unit pulses in a vector of dimension N, where two pulses at the same position are required to have the same sign. Thus the codebook includes all integer codevectors y of dimension N that satisfy <figure align="center"> <artwork align="center"><![CDATA[ N-1 __ \ abs(y[j]) = K . /_ j=0 ]]></artwork> </figure> Unlike regular PVQ, SILK uses a variable-length, rather than fixed-length, encoding. This encoding is better suited to the more Gaussian-like distribution of the coefficient magnitudes and the non-uniform distribution of their signs (caused by the quantization offset described below). SILK also handles large codebooks by coding the least significant bits (LSb's) of each coefficient directly. This adds a small coding efficiency loss, but greatly reduces the computation time and ROM size required for decoding, as implemented in silk_decode_pulses() (decode_pulses.c). </t> <t> SILK fixes the dimension of the codebook to N = 16. The excitation is made up of a number of "shell blocks", each 16 samples in size. <xref target="silk_shell_block_table"/> lists the number of shell blocks required for a SILK frame for each possible audio bandwidth and frame size. 10 ms MB frames nominally contain 120 samples (10 ms at 12 kHz), which is not a multiple of 16. This is handled by coding 8 shell blocks (128 samples) and discarding the final 8 samples of the last block. The decoder contains no special case that prevents an encoder from placing pulses in these samples, and they must be correctly parsed from the bitstream if present, but they are otherwise ignored. </t> <texttable anchor="silk_shell_block_table" title="Number of Shell Blocks Per SILK Frame"> <ttcol>Audio Bandwidth</ttcol> <ttcol>Frame Size</ttcol> <ttcol align="right">Number of Shell Blocks</ttcol> <c>NB</c> <c>10 ms</c> <c>5</c> <c>MB</c> <c>10 ms</c> <c>8</c> <c>WB</c> <c>10 ms</c> <c>10</c> <c>NB</c> <c>20 ms</c> <c>10</c> <c>MB</c> <c>20 ms</c> <c>15</c> <c>WB</c> <c>20 ms</c> <c>20</c> </texttable> <section anchor="silk_rate_level" title="Rate Level"> <t> The first symbol in the excitation is a "rate level", which is an index from 0 to 8, inclusive, coded using the PDF in <xref target="silk_rate_level_pdfs"/> corresponding to the signal type of the current frame (from <xref target="silk_frame_type"/>). The rate level selects the PDF used to decode the number of pulses in the individual shell blocks. It does not directly convey any information about the bitrate or the number of pulses itself, but merely changes the probability of the symbols in <xref target="silk_pulse_counts"/>. Level 0 provides a more efficient encoding at low rates generally, and level 8 provides a more efficient encoding at high rates generally, though the most efficient level for a particular SILK frame may depend on the exact distribution of the coded symbols. An encoder should, but is not required to, use the most efficient rate level. </t> <texttable anchor="silk_rate_level_pdfs" title="PDFs for the Rate Level"> <ttcol>Signal Type</ttcol> <ttcol>PDF</ttcol> <c>Inactive or Unvoiced</c> <c>{15, 51, 12, 46, 45, 13, 33, 27, 14}/256</c> <c>Voiced</c> <c>{33, 30, 36, 17, 34, 49, 18, 21, 18}/256</c> </texttable> </section> <section anchor="silk_pulse_counts" title="Pulses Per Shell Block"> <t> The total number of pulses in each of the shell blocks follows the rate level. The pulse counts for all of the shell blocks are coded consecutively, before the content of any of the blocks. Each block may have anywhere from 0 to 16 pulses, inclusive, coded using the 18-entry PDF in <xref target="silk_pulse_count_pdfs"/> corresponding to the rate level from <xref target="silk_rate_level"/>. The special value 17 indicates that this block has one or more additional LSb's to decode for each coefficient. If the decoder encounters this value, it decodes another value for the actual pulse count of the block, but uses the PDF corresponding to the special rate level 9 instead of the normal rate level. This process repeats until the decoder reads a value less than 17, and it then sets the number of extra LSb's used to the number of 17's decoded for that block. If it reads the value 17 ten times, then the next iteration uses the special rate level 10 instead of 9. The probability of decoding a 17 when using the PDF for rate level 10 is zero, ensuring that the number of LSb's for a block will not exceed 10. The cumulative distribution for rate level 10 is just a shifted version of that for 9 and thus does not require any additional storage. </t> <texttable anchor="silk_pulse_count_pdfs" title="PDFs for the Pulse Count"> <ttcol>Rate Level</ttcol> <ttcol>PDF</ttcol> <c>0</c> <c>{131, 74, 25, 8, 3, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}/256</c> <c>1</c> <c>{58, 93, 60, 23, 7, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}/256</c> <c>2</c> <c>{43, 51, 46, 33, 24, 16, 11, 8, 6, 3, 3, 3, 2, 1, 1, 2, 1, 2}/256</c> <c>3</c> <c>{17, 52, 71, 57, 31, 12, 5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}/256</c> <c>4</c> <c>{6, 21, 41, 53, 49, 35, 21, 11, 6, 3, 2, 2, 1, 1, 1, 1, 1, 1}/256</c> <c>5</c> <c>{7, 14, 22, 28, 29, 28, 25, 20, 17, 13, 11, 9, 7, 5, 4, 4, 3, 10}/256</c> <c>6</c> <c>{2, 5, 14, 29, 42, 46, 41, 31, 19, 11, 6, 3, 2, 1, 1, 1, 1, 1}/256</c> <c>7</c> <c>{1, 2, 4, 10, 19, 29, 35, 37, 34, 28, 20, 14, 8, 5, 4, 2, 2, 2}/256</c> <c>8</c> <c>{1, 2, 2, 5, 9, 14, 20, 24, 27, 28, 26, 23, 20, 15, 11, 8, 6, 15}/256</c> <c>9</c> <c>{1, 1, 1, 6, 27, 58, 56, 39, 25, 14, 10, 6, 3, 3, 2, 1, 1, 2}/256</c> <c>10</c> <c>{2, 1, 6, 27, 58, 56, 39, 25, 14, 10, 6, 3, 3, 2, 1, 1, 2, 0}/256</c> </texttable> </section> <section anchor="silk_pulse_locations" title="Pulse Location Decoding"> <t> The locations of the pulses in each shell block follow the pulse counts, as decoded by silk_shell_decoder() (shell_coder.c). As with the pulse counts, these locations are coded for all the shell blocks before any of the remaining information for each block. Unlike many other codecs, SILK places no restriction on the distribution of pulses within a shell block. All of the pulses may be placed in a single location, or each one in a unique location, or anything in between. </t> <t> The location of pulses is coded by recursively partitioning each block into halves, and coding how many pulses fall on the left side of the split. All remaining pulses must fall on the right side of the split. The process then recurses into the left half, and after that returns, the right half (preorder traversal). The PDF to use is chosen by the size of the current partition (16, 8, 4, or 2) and the number of pulses in the partition (1 to 16, inclusive). Tables <xref format="counter" target="silk_shell_code3_pdfs"/> through <xref format="counter" target="silk_shell_code0_pdfs"/> list the PDFs used for each partition size and pulse count. This process skips partitions without any pulses, i.e., where the initial pulse count from <xref target="silk_pulse_counts"/> was zero, or where the split in the prior level indicated that all of the pulses fell on the other side. These partitions have nothing to code, so they require no PDF. </t> <texttable anchor="silk_shell_code3_pdfs" title="PDFs for Pulse Count Split, 16 Sample Partitions"> <ttcol>Pulse Count</ttcol> <ttcol>PDF</ttcol> <c>1</c> <c>{126, 130}/256</c> <c>2</c> <c>{56, 142, 58}/256</c> <c>3</c> <c>{25, 101, 104, 26}/256</c> <c>4</c> <c>{12, 60, 108, 64, 12}/256</c> <c>5</c> <c>{7, 35, 84, 87, 37, 6}/256</c> <c>6</c> <c>{4, 20, 59, 86, 63, 21, 3}/256</c> <c>7</c> <c>{3, 12, 38, 72, 75, 42, 12, 2}/256</c> <c>8</c> <c>{2, 8, 25, 54, 73, 59, 27, 7, 1}/256</c> <c>9</c> <c>{2, 5, 17, 39, 63, 65, 42, 18, 4, 1}/256</c> <c>10</c> <c>{1, 4, 12, 28, 49, 63, 54, 30, 11, 3, 1}/256</c> <c>11</c> <c>{1, 4, 8, 20, 37, 55, 57, 41, 22, 8, 2, 1}/256</c> <c>12</c> <c>{1, 3, 7, 15, 28, 44, 53, 48, 33, 16, 6, 1, 1}/256</c> <c>13</c> <c>{1, 2, 6, 12, 21, 35, 47, 48, 40, 25, 12, 5, 1, 1}/256</c> <c>14</c> <c>{1, 1, 4, 10, 17, 27, 37, 47, 43, 33, 21, 9, 4, 1, 1}/256</c> <c>15</c> <c>{1, 1, 1, 8, 14, 22, 33, 40, 43, 38, 28, 16, 8, 1, 1, 1}/256</c> <c>16</c> <c>{1, 1, 1, 1, 13, 18, 27, 36, 41, 41, 34, 24, 14, 1, 1, 1, 1}/256</c> </texttable> <texttable anchor="silk_shell_code2_pdfs" title="PDFs for Pulse Count Split, 8 Sample Partitions"> <ttcol>Pulse Count</ttcol> <ttcol>PDF</ttcol> <c>1</c> <c>{127, 129}/256</c> <c>2</c> <c>{53, 149, 54}/256</c> <c>3</c> <c>{22, 105, 106, 23}/256</c> <c>4</c> <c>{11, 61, 111, 63, 10}/256</c> <c>5</c> <c>{6, 35, 86, 88, 36, 5}/256</c> <c>6</c> <c>{4, 20, 59, 87, 62, 21, 3}/256</c> <c>7</c> <c>{3, 13, 40, 71, 73, 41, 13, 2}/256</c> <c>8</c> <c>{3, 9, 27, 53, 70, 56, 28, 9, 1}/256</c> <c>9</c> <c>{3, 8, 19, 37, 57, 61, 44, 20, 6, 1}/256</c> <c>10</c> <c>{3, 7, 15, 28, 44, 54, 49, 33, 17, 5, 1}/256</c> <c>11</c> <c>{1, 7, 13, 22, 34, 46, 48, 38, 28, 14, 4, 1}/256</c> <c>12</c> <c>{1, 1, 11, 22, 27, 35, 42, 47, 33, 25, 10, 1, 1}/256</c> <c>13</c> <c>{1, 1, 6, 14, 26, 37, 43, 43, 37, 26, 14, 6, 1, 1}/256</c> <c>14</c> <c>{1, 1, 4, 10, 20, 31, 40, 42, 40, 31, 20, 10, 4, 1, 1}/256</c> <c>15</c> <c>{1, 1, 3, 8, 16, 26, 35, 38, 38, 35, 26, 16, 8, 3, 1, 1}/256</c> <c>16</c> <c>{1, 1, 2, 6, 12, 21, 30, 36, 38, 36, 30, 21, 12, 6, 2, 1, 1}/256</c> </texttable> <texttable anchor="silk_shell_code1_pdfs" title="PDFs for Pulse Count Split, 4 Sample Partitions"> <ttcol>Pulse Count</ttcol> <ttcol>PDF</ttcol> <c>1</c> <c>{127, 129}/256</c> <c>2</c> <c>{49, 157, 50}/256</c> <c>3</c> <c>{20, 107, 109, 20}/256</c> <c>4</c> <c>{11, 60, 113, 62, 10}/256</c> <c>5</c> <c>{7, 36, 84, 87, 36, 6}/256</c> <c>6</c> <c>{6, 24, 57, 82, 60, 23, 4}/256</c> <c>7</c> <c>{5, 18, 39, 64, 68, 42, 16, 4}/256</c> <c>8</c> <c>{6, 14, 29, 47, 61, 52, 30, 14, 3}/256</c> <c>9</c> <c>{1, 15, 23, 35, 51, 50, 40, 30, 10, 1}/256</c> <c>10</c> <c>{1, 1, 21, 32, 42, 52, 46, 41, 18, 1, 1}/256</c> <c>11</c> <c>{1, 6, 16, 27, 36, 42, 42, 36, 27, 16, 6, 1}/256</c> <c>12</c> <c>{1, 5, 12, 21, 31, 38, 40, 38, 31, 21, 12, 5, 1}/256</c> <c>13</c> <c>{1, 3, 9, 17, 26, 34, 38, 38, 34, 26, 17, 9, 3, 1}/256</c> <c>14</c> <c>{1, 3, 7, 14, 22, 29, 34, 36, 34, 29, 22, 14, 7, 3, 1}/256</c> <c>15</c> <c>{1, 2, 5, 11, 18, 25, 31, 35, 35, 31, 25, 18, 11, 5, 2, 1}/256</c> <c>16</c> <c>{1, 1, 4, 9, 15, 21, 28, 32, 34, 32, 28, 21, 15, 9, 4, 1, 1}/256</c> </texttable> <texttable anchor="silk_shell_code0_pdfs" title="PDFs for Pulse Count Split, 2 Sample Partitions"> <ttcol>Pulse Count</ttcol> <ttcol>PDF</ttcol> <c>1</c> <c>{128, 128}/256</c> <c>2</c> <c>{42, 172, 42}/256</c> <c>3</c> <c>{21, 107, 107, 21}/256</c> <c>4</c> <c>{12, 60, 112, 61, 11}/256</c> <c>5</c> <c>{8, 34, 86, 86, 35, 7}/256</c> <c>6</c> <c>{8, 23, 55, 90, 55, 20, 5}/256</c> <c>7</c> <c>{5, 15, 38, 72, 72, 36, 15, 3}/256</c> <c>8</c> <c>{6, 12, 27, 52, 77, 47, 20, 10, 5}/256</c> <c>9</c> <c>{6, 19, 28, 35, 40, 40, 35, 28, 19, 6}/256</c> <c>10</c> <c>{4, 14, 22, 31, 37, 40, 37, 31, 22, 14, 4}/256</c> <c>11</c> <c>{3, 10, 18, 26, 33, 38, 38, 33, 26, 18, 10, 3}/256</c> <c>12</c> <c>{2, 8, 13, 21, 29, 36, 38, 36, 29, 21, 13, 8, 2}/256</c> <c>13</c> <c>{1, 5, 10, 17, 25, 32, 38, 38, 32, 25, 17, 10, 5, 1}/256</c> <c>14</c> <c>{1, 4, 7, 13, 21, 29, 35, 36, 35, 29, 21, 13, 7, 4, 1}/256</c> <c>15</c> <c>{1, 2, 5, 10, 17, 25, 32, 36, 36, 32, 25, 17, 10, 5, 2, 1}/256</c> <c>16</c> <c>{1, 2, 4, 7, 13, 21, 28, 34, 36, 34, 28, 21, 13, 7, 4, 2, 1}/256</c> </texttable> </section> <section anchor="silk_shell_lsb" title="LSb Decoding"> <t> After the decoder reads the pulse locations for all blocks, it reads the LSb's (if any) for each block in turn. Inside each block, it reads all the LSb's for each coefficient in turn, even those where no pulses were allocated, before proceeding to the next one. They are coded from most significant to least significant, and they all use the PDF in <xref target="silk_shell_lsb_pdf"/>. </t> <texttable anchor="silk_shell_lsb_pdf" title="PDF for Excitation LSb's"> <ttcol>PDF</ttcol> <c>{136, 120}/256</c> </texttable> <t> The number of LSb's read for each coefficient in a block is determined in <xref target="silk_pulse_counts"/>. The magnitude of the coefficient is initially equal to the number of pulses placed at that location in <xref target="silk_pulse_locations"/>. As each LSb is decoded, the magnitude is doubled, and then the value of the LSb added to it, to obtain an updated magnitude. </t> </section> <section anchor="silk_signs" title="Sign Decoding"> <t> After decoding the pulse locations and the LSb's, the decoder knows the magnitude of each coefficient in the excitation. It then decodes a sign for all coefficients with a non-zero magnitude, using one of the PDFs from <xref target="silk_sign_pdfs"/>. If the value decoded is 0, then the coefficient magnitude is negated. Otherwise, it remains positive. </t> <t> The decoder chooses the PDF for the sign based on the signal type and quantization offset type (from <xref target="silk_frame_type"/>) and the number of pulses in the block (from <xref target="silk_pulse_counts"/>). The number of pulses in the block does not take into account any LSb's. Most PDFs are skewed towards negative signs because of the quantizaton offset, but the PDFs for zero pulses are highly skewed towards positive signs. If a block contains many positive coefficients, it is sometimes beneficial to code it solely using LSb's (i.e., with zero pulses), since the encoder may be able to save enough bits on the signs to justify the less efficient coefficient magnitude encoding. </t> <texttable anchor="silk_sign_pdfs" title="PDFs for Excitation Signs"> <ttcol>Signal Type</ttcol> <ttcol>Quantization Offset Type</ttcol> <ttcol>Pulse Count</ttcol> <ttcol>PDF</ttcol> <c>Inactive</c> <c>Low</c> <c>0</c> <c>{2, 254}/256</c> <c>Inactive</c> <c>Low</c> <c>1</c> <c>{207, 49}/256</c> <c>Inactive</c> <c>Low</c> <c>2</c> <c>{189, 67}/256</c> <c>Inactive</c> <c>Low</c> <c>3</c> <c>{179, 77}/256</c> <c>Inactive</c> <c>Low</c> <c>4</c> <c>{174, 82}/256</c> <c>Inactive</c> <c>Low</c> <c>5</c> <c>{163, 93}/256</c> <c>Inactive</c> <c>Low</c> <c>6 or more</c> <c>{157, 99}/256</c> <c>Inactive</c> <c>High</c> <c>0</c> <c>{58, 198}/256</c> <c>Inactive</c> <c>High</c> <c>1</c> <c>{245, 11}/256</c> <c>Inactive</c> <c>High</c> <c>2</c> <c>{238, 18}/256</c> <c>Inactive</c> <c>High</c> <c>3</c> <c>{232, 24}/256</c> <c>Inactive</c> <c>High</c> <c>4</c> <c>{225, 31}/256</c> <c>Inactive</c> <c>High</c> <c>5</c> <c>{220, 36}/256</c> <c>Inactive</c> <c>High</c> <c>6 or more</c> <c>{211, 45}/256</c> <c>Unvoiced</c> <c>Low</c> <c>0</c> <c>{1, 255}/256</c> <c>Unvoiced</c> <c>Low</c> <c>1</c> <c>{210, 46}/256</c> <c>Unvoiced</c> <c>Low</c> <c>2</c> <c>{190, 66}/256</c> <c>Unvoiced</c> <c>Low</c> <c>3</c> <c>{178, 78}/256</c> <c>Unvoiced</c> <c>Low</c> <c>4</c> <c>{169, 87}/256</c> <c>Unvoiced</c> <c>Low</c> <c>5</c> <c>{162, 94}/256</c> <c>Unvoiced</c> <c>Low</c> <c>6 or more</c> <c>{152, 104}/256</c> <c>Unvoiced</c> <c>High</c> <c>0</c> <c>{48, 208}/256</c> <c>Unvoiced</c> <c>High</c> <c>1</c> <c>{242, 14}/256</c> <c>Unvoiced</c> <c>High</c> <c>2</c> <c>{235, 21}/256</c> <c>Unvoiced</c> <c>High</c> <c>3</c> <c>{224, 32}/256</c> <c>Unvoiced</c> <c>High</c> <c>4</c> <c>{214, 42}/256</c> <c>Unvoiced</c> <c>High</c> <c>5</c> <c>{205, 51}/256</c> <c>Unvoiced</c> <c>High</c> <c>6 or more</c> <c>{190, 66}/256</c> <c>Voiced</c> <c>Low</c> <c>0</c> <c>{1, 255}/256</c> <c>Voiced</c> <c>Low</c> <c>1</c> <c>{162, 94}/256</c> <c>Voiced</c> <c>Low</c> <c>2</c> <c>{152, 104}/256</c> <c>Voiced</c> <c>Low</c> <c>3</c> <c>{147, 109}/256</c> <c>Voiced</c> <c>Low</c> <c>4</c> <c>{144, 112}/256</c> <c>Voiced</c> <c>Low</c> <c>5</c> <c>{141, 115}/256</c> <c>Voiced</c> <c>Low</c> <c>6 or more</c> <c>{138, 118}/256</c> <c>Voiced</c> <c>High</c> <c>0</c> <c>{8, 248}/256</c> <c>Voiced</c> <c>High</c> <c>1</c> <c>{203, 53}/256</c> <c>Voiced</c> <c>High</c> <c>2</c> <c>{187, 69}/256</c> <c>Voiced</c> <c>High</c> <c>3</c> <c>{176, 80}/256</c> <c>Voiced</c> <c>High</c> <c>4</c> <c>{168, 88}/256</c> <c>Voiced</c> <c>High</c> <c>5</c> <c>{161, 95}/256</c> <c>Voiced</c> <c>High</c> <c>6 or more</c> <c>{154, 102}/256</c> </texttable> </section> <section anchor="silk_excitation_reconstruction" title="Reconstructing the Excitation"> <t> After the signs have been read, there is enough information to reconstruct the complete excitation signal. This requires adding a constant quantization offset to each non-zero sample, and then pseudorandomly inverting and offsetting every sample. The constant quantization offset varies depending on the signal type and quantization offset type (see <xref target="silk_frame_type"/>). </t> <texttable anchor="silk_quantization_offsets" title="Excitation Quantization Offsets"> <ttcol align="left">Signal Type</ttcol> <ttcol align="left">Quantization Offset Type</ttcol> <ttcol align="right">Quantization Offset (Q10)</ttcol> <c>Inactive</c> <c>Low</c> <c>100</c> <c>Inactive</c> <c>High</c> <c>240</c> <c>Unvoiced</c> <c>Low</c> <c>100</c> <c>Unvoiced</c> <c>High</c> <c>240</c> <c>Voiced</c> <c>Low</c> <c>32</c> <c>Voiced</c> <c>High</c> <c>100</c> </texttable> <t> Let e_raw[i] be the raw excitation value at position i, with a magnitude composed of the pulses at that location (see <xref target="silk_pulse_locations"/>) combined with any additional LSb's (see <xref target="silk_shell_lsb"/>), and with the corresponding sign decoded in <xref target="silk_signs"/>. Additionally, let seed be the current pseudorandom seed, which is initialized to the value decoded from <xref target="silk_seed"/> for the first sample in the current SILK frame, and updated for each subsequent sample according to the procedure below. Finally, let offset_Q10 be the quantization offset from <xref target="silk_quantization_offsets"/>. Then the following procedure produces the final reconstructed excitation value, e_Q10[i]: <figure align="center"> <artwork align="center"><![CDATA[ e_Q10[i] = (e_raw[i] << 10) - sign(e_raw[i])*offset_Q10; seed = (196314165*seed + 907633515) & 0xFFFFFFFF; e_Q10[i] = (seed & 0x80000000) ? -(e_Q10[i] + 1) : e_Q10[i]; seed = (seed + e_raw[i]) & 0xFFFFFFFF; ]]></artwork> </figure> When e_raw[i] is zero, sign() returns 0 by the definition in <xref target="sign"/>, implying that no quantization offset gets added. The final e_Q10[i] value may require more than 16 bits per sample, but will not require more than 32. </t> </section> </section> <section anchor="silk_frame_reconstruction" title="SILK Frame Reconstruction"/> <section anchor="silk_ltp_synthesis" title="LTP Synthesis"> <t> For voiced speech, the excitation signal e(n) is input to an LTP synthesis filter that recreates the long-term correlation removed in the LTP analysis filter and generates an LPC excitation signal e_LPC(n), according to <figure align="center"> <artwork align="center"><![CDATA[ d __ e_LPC(n) = e(n) + \ e_LPC(n - L - i) * b_i, /_ i=-d ]]></artwork> </figure> using the pitch lag L, and the decoded LTP coefficients b_i. The number of LTP coefficients is 5, and thus d = 2. For unvoiced speech, the output signal is simply a copy of the excitation signal, i.e., e_LPC(n) = e(n). </t> </section> <section anchor="silk_lpc_synthesis" title='LPC Synthesis'> <t> In a similar manner, the short-term correlation that was removed in the LPC analysis filter is recreated in the LPC synthesis filter. The LPC excitation signal e_LPC(n) is filtered using the LTP coefficients a_i, according to <figure align="center"> <artwork align="center"><![CDATA[ d_LPC __ y(n) = e_LPC(n) + \ y(n - i) * a_i, /_ i=1 ]]></artwork> </figure> where d_LPC is the LPC synthesis filter order, and y(n) is the decoded output signal. </t> </section> </section> </section> <section title="CELT Decoder"> <t> An overview of the decoder is given in <xref target="celt-decoder-overview"/>. </t> <figure anchor="celt-decoder-overview" title="Structure of the CELT decoder"> <artwork align="center"><![CDATA[ +---------+ | Coarse | +->| decoder |----+ | +---------+ | | | | +---------+ v | | Fine | +---+ +->| decoder |->| + | | +---------+ +---+ | ^ | +---------+ | | | | Range | | +----------+ v | Decoder |-+ | Bit | +-----+ +---------+ | |Allocation| | 2^x | | +----------+ +-----+ | | | | v v +--------+ | +---------+ +---+ +-------+ | pitch | +->| PVQ |->| * |->| IMDCT |->| post- |---> | | decoder | +---+ +-------+ | filter | | +---------+ +--------+ | ^ +--------------------------------------+ ]]></artwork> </figure> <t> The decoder is based on the following symbols and sets of symbols: </t> <texttable anchor="celt_symbols"> <ttcol align="center">Symbol(s)</ttcol> <ttcol align="center">PDF</ttcol> <ttcol align="center">Condition</ttcol> <c>silence</c> <c>{32767, 1}/32768</c> <c></c> <c>post-filter</c> <c>{1, 1}/2</c> <c></c> <c>octave</c> <c>uniform (6)</c><c>post-filter</c> <c>period</c> <c>raw bits (4+octave)</c><c>post-filter</c> <c>gain</c> <c>raw bits (3)</c><c>post-filter</c> <c>tapset</c> <c>{2, 1, 1}/4</c><c>post-filter</c> <c>transient</c> <c>{7, 1}/8</c><c></c> <c>intra</c> <c>{7, 1}/8</c><c></c> <c>coarse energy</c><c><xref target="energy-decoding"/></c><c></c> <c>tf_change</c> <c><xref target="transient-decoding"/></c><c></c> <c>tf_select</c> <c>{1, 1}/2</c><c><xref target="transient-decoding"/></c> <c>spread</c> <c>{7, 2, 21, 2}/32</c><c></c> <c>dyn. alloc.</c> <c><xref target="allocation"/></c><c></c> <c>alloc. trim</c> <c>{2, 2, 5, 10, 22, 46, 22, 10, 5, 2, 2}/128</c><c></c> <c>skip</c> <c>{1, 1}/2</c><c><xref target="allocation"/></c> <c>intensity</c> <c>uniform</c><c><xref target="allocation"/></c> <c>dual</c> <c>{1, 1}/2</c><c></c> <c>fine energy</c> <c><xref target="energy-decoding"/></c><c></c> <c>residual</c> <c><xref target="PVQ-decoder"/></c><c></c> <c>anti-collapse</c><c>{1, 1}/2</c><c><xref target="anti-collapse"/></c> <c>finalize</c> <c><xref target="energy-decoding"/></c><c></c> <postamble>Order of the symbols in the CELT section of the bitstream.</postamble> </texttable> <t> The decoder extracts information from the range-coded bitstream in the order described in the figure above. In some circumstances, it is possible for a decoded value to be out of range due to a very small amount of redundancy in the encoding of large integers by the range coder. In that case, the decoder should assume there has been an error in the coding, decoding, or transmission and SHOULD take measures to conceal the error and/or report to the application that a problem has occurred. </t> <section anchor="transient-decoding" title="Transient Decoding"> <t> The "transient" flag encoded in the bitstream has a probability of 1/8. When it is set, then the MDCT coefficients represent multiple short MDCTs in the frame. When not set, the coefficients represent a single long MDCT for the frame. In addition to the global transient flag is a per-band binary flag to change the time-frequency (tf) resolution independently in each band. The change in tf resolution is defined in tf_select_table[][] in celt.c and depends on the frame size, whether the transient flag is set, and the value of tf_select. The tf_select flag uses a 1/2 probability, but is only decoded if it can have an impact on the result knowing the value of all per-band tf_change flags. </t> </section> <section anchor="energy-decoding" title="Energy Envelope Decoding"> <t> It is important to quantize the energy with sufficient resolution because any energy quantization error cannot be compensated for at a later stage. Regardless of the resolution used for encoding the shape of a band, it is perceptually important to preserve the energy in each band. CELT uses a three-step coarse-fine-fine strategy for encoding the energy in the base-2 log domain, as implemented in quant_bands.c</t> <section anchor="coarse-energy-decoding" title="Coarse energy decoding"> <t> Coarse quantization of the energy uses a fixed resolution of 6 dB (integer part of base-2 log). To minimize the bitrate, prediction is applied both in time (using the previous frame) and in frequency (using the previous bands). The part of the prediction that is based on the previous frame can be disabled, creating an "intra" frame where the energy is coded without reference to prior frames. The decoder first reads the intra flag to determine what prediction is used. The 2-D z-transform of the prediction filter is: <figure align="center"> <artwork align="center"><![CDATA[ -1 -1 (1 - alpha*z_l )*(1 - z_b ) A(z_l, z_b) = ----------------------------- -1 1 - beta*z_b ]]></artwork> </figure> where b is the band index and l is the frame index. The prediction coefficients applied depend on the frame size in use when not using intra energy and are alpha=0, beta=4915/32768 when using intra energy. The time-domain prediction is based on the final fine quantization of the previous frame, while the frequency domain (within the current frame) prediction is based on coarse quantization only (because the fine quantization has not been computed yet). The prediction is clamped internally so that fixed point implementations with limited dynamic range do not suffer desynchronization. We approximate the ideal probability distribution of the prediction error using a Laplace distribution with separate parameters for each frame size in intra- and inter-frame modes. The coarse energy quantization is performed by unquant_coarse_energy() and unquant_coarse_energy_impl() (quant_bands.c). The encoding of the Laplace-distributed values is implemented in ec_laplace_decode() (laplace.c). </t> </section> <section anchor="fine-energy-decoding" title="Fine energy quantization"> <t> The number of bits assigned to fine energy quantization in each band is determined by the bit allocation computation described in <xref target="allocation"></xref>. Let B_i be the number of fine energy bits for band i; the refinement is an integer f in the range [0,2**B_i-1]. The mapping between f and the correction applied to the coarse energy is equal to (f+1/2)/2**B_i - 1/2. Fine energy quantization is implemented in quant_fine_energy() (quant_bands.c). </t> <t> When some bits are left "unused" after all other flags have been decoded, these bits are assigned to a "final" step of fine allocation. In effect, these bits are used to add one extra fine energy bit per band per channel. The allocation process determines two "priorities" for the final fine bits. Any remaining bits are first assigned only to bands of priority 0, starting from band 0 and going up. If all bands of priority 0 have received one bit per channel, then bands of priority 1 are assigned an extra bit per channel, starting from band 0. If any bits are left after this, they are left unused. This is implemented in unquant_energy_finalise() (quant_bands.c). </t> </section> <!-- fine energy --> </section> <!-- Energy decode --> <section anchor="allocation" title="Bit allocation"> <t>Many codecs transmit significant amounts of side information for the purpose of controlling bit allocation within a frame. Often this side information controls bit usage indirectly and must be carefully selected to achieve the desired rate constraints.</t> <t>The band-energy normalized structure of Opus MDCT mode ensures that a constant bit allocation for the shape content of a band will result in a roughly constant tone to noise ratio, which provides for fairly consistent perceptual performance. The effectiveness of this approach is the result of two factors: that the band energy, which is understood to be perceptually important on its own, is always preserved regardless of the shape precision, and because the constant tone-to-noise ratio implies a constant intra-band noise to masking ratio. Intra-band masking is the strongest of the perceptual masking effects. This structure means that the ideal allocation is more consistent from frame to frame than it is for other codecs without an equivalent structure.</t> <t>Because the bit allocation is used to drive the decoding of the range-coder stream, it MUST be recovered exactly so that identical coding decisions are made in the encoder and decoder. Any deviation from the reference's resulting bit allocation will result in corrupted output, though implementers are free to implement the procedure in any way which produces identical results.</t> <t>Because all of the information required to decode a frame must be derived from that frame alone in order to retain robustness to packet loss, the overhead of explicitly signaling the allocation would be considerable, especially for low-latency (small frame size) applications, even though the allocation is relatively static.</t> <t>For this reason, in the MDCT mode Opus uses a primarily implicit bit allocation. The available bitstream capacity is known in advance to both the encoder and decoder without additional signaling, ultimately from the packet sizes expressed by a higher-level protocol. Using this information the codec interpolates an allocation from a hard-coded table.</t> <t>While the band-energy structure effectively models intra-band masking, it ignores the weaker inter-band masking, band-temporal masking, and other less significant perceptual effects. While these effects can often be ignored, they can become significant for particular samples. One mechanism available to encoders would be to simply increase the overall rate for these frames, but this is not possible in a constant rate mode and can be fairly inefficient. As a result three explicitly signaled mechanisms are provided to alter the implicit allocation:</t> <t> <list style="symbols"> <t>Band boost</t> <t>Allocation trim</t> <t>Band skipping</t> </list> </t> <t>The first of these mechanisms, band boost, allows an encoder to boost the allocation in specific bands. The second, allocation trim, works by biasing the overall allocation towards higher or lower frequency bands. The third, band skipping, selects which low-precision high frequency bands will be allocated no shape bits at all.</t> <t>In stereo mode there are two additional parameters potentially coded as part of the allocation procedure: a parameter to allow the selective elimination of allocation for the 'side' in jointly coded bands, and a flag to deactivate joint coding. These values are not signaled if they would be meaningless in the overall context of the allocation.</t> <t>Because every signaled adjustment increases overhead and implementation complexity, none were included speculatively: the reference encoder makes use of all of these mechanisms. While the decision logic in the reference was found to be effective enough to justify the overhead and complexity, further analysis techniques may be discovered which increase the effectiveness of these parameters. As with other signaled parameters, an encoder is free to choose the values in any manner, but unless a technique is known to deliver superior perceptual results the methods used by the reference implementation should be used.</t> <t>The allocation process consists of the following steps: determining the per-band maximum allocation vector, decoding the boosts, decoding the tilt, determining the remaining capacity of the frame, searching the mode table for the entry nearest but not exceeding the available space (subject to the tilt, boosts, band maximums, and band minimums), linear interpolation, reallocation of unused bits with concurrent skip decoding, determination of the fine-energy vs. shape split, and final reallocation. This process results in a per-band shape allocation (in 1/8th bit units), a per-band fine-energy allocation (in 1 bit per channel units), a set of band priorities for controlling the use of remaining bits at the end of the frame, and a remaining balance of unallocated space, which is usually zero except at very high rates.</t> <t>The maximum allocation vector is an approximation of the maximum space that can be used by each band for a given mode. The value is approximate because the shape encoding is variable rate (due to entropy coding of splitting parameters). Setting the maximum too low reduces the maximum achievable quality in a band while setting it too high may result in waste: bitstream capacity available at the end of the frame which can not be put to any use. The maximums specified by the codec reflect the average maximum. In the reference the maximums are provided in partially computed form, in order to fit in less memory as a static table (XXX cache.caps). Implementations are expected to simply use the same table data, but the procedure for generating this table is included in rate.c as part of compute_pulse_cache().</t> <t>To convert the values in cache.caps into the actual maximums: first set nbBands to the maximum number of bands for this mode, and stereo to zero if stereo is not in use and one otherwise. For each band set N to the number of MDCT bins covered by the band (for one channel), set LM to the shift value for the frame size (e.g. 0 for 120, 1 for 240, 3 for 480), then set i to nbBands*(2*LM+stereo). Then set the maximum for the band to the i-th index of cache.caps + 64 and multiply by the number of channels in the current frame (one or two) and by N, then divide the result by 4 using truncating integer division. The resulting vector will be called cap[]. The elements fit in signed 16-bit integers but do not fit in 8 bits. This procedure is implemented in the reference in the function init_caps() in celt.c. </t> <t>The band boosts are represented by a series of binary symbols which are coded with very low probability. Each band can potentially be boosted multiple times, subject to the frame actually having enough room to obey the boost and having enough room to code the boost symbol. The default coding cost for a boost starts out at six bits, but subsequent boosts in a band cost only a single bit and every time a band is boosted the initial cost is reduced (down to a minimum of two). Since the initial cost of coding a boost is 6 bits, the coding cost of the boost symbols when completely unused is 0.48 bits/frame for a 21 band mode (21*-log2(1-1/2**6)).</t> <t>To decode the band boosts: First set 'dynalloc_logp' to 6, the initial amount of storage required to signal a boost in bits, 'total_bits' to the size of the frame in 8th bits, 'total_boost' to zero, and 'tell' to the total number of 8th bits decoded so far. For each band from the coding start (0 normally, but 17 in hybrid mode) to the coding end (which changes depending on the signaled bandwidth): set 'width' to the number of MDCT bins in this band for all channels. Take the larger of width and 64, then the minimum of that value and the width times eight and set 'quanta' to the result. This represents a boost step size of six bits subject to limits of 1/bit/sample and 1/8th bit/sample. Set 'boost' to zero and 'dynalloc_loop_logp' to dynalloc_logp. While dynalloc_loop_log (the current worst case symbol cost) in 8th bits plus tell is less than total_bits plus total_boost and boost is less than cap[] for this band: Decode a bit from the bitstream with a with dynalloc_loop_logp as the cost of a one, update tell to reflect the current used capacity, if the decoded value is zero break the loop otherwise add quanta to boost and total_boost, subtract quanta from total_bits, and set dynalloc_loop_log to 1. When the while loop finishes boost contains the boost for this band. If boost is non-zero and dynalloc_logp is greater than 2, decrease dynalloc_logp. Once this process has been executed on all bands, the band boosts have been decoded. This procedure is implemented around line 2352 of celt.c.</t> <t>At very low rates it is possible that there won't be enough available space to execute the inner loop even once. In these cases band boost is not possible but its overhead is completely eliminated. Because of the high cost of band boost when activated, a reasonable encoder should not be using it at very low rates. The reference implements its dynalloc decision logic around line 1269 of celt.c.</t> <t>The allocation trim is a integer value from 0-10. The default value of 5 indicates no trim. The trim parameter is entropy coded in order to lower the coding cost of less extreme adjustments. Values lower than 5 bias the allocation towards lower frequencies and values above 5 bias it towards higher frequencies. Like other signaled parameters, signaling of the trim is gated so that it is not included if there is insufficient space available in the bitstream. To decode the trim, first set the trim value to 5, then iff the count of decoded 8th bits so far (ec_tell_frac) plus 48 (6 bits) is less than or equal to the total frame size in 8th bits minus total_boost (a product of the above band boost procedure), decode the trim value using the inverse CDF {127, 126, 124, 119, 109, 87, 41, 19, 9, 4, 2, 0}.</t> <t>For 10 ms and 20 ms frames using short blocks and that have at least LM+2 bits left prior to the allocation process, then one anti-collapse bit is reserved in the allocation process so it can be decoded later. Following the the anti-collapse reservation, one bit is reserved for skip if available.</t> <t>For stereo frames, bits are reserved for intensity stereo and for dual stereo. Intensity stereo requires ilog2(end-start) bits. Those bits are reserved if there is enough bits left. Following this, one bit is reserved for dual stereo if available.</t> <t>The allocation computation begins by setting up some initial conditions. 'total' is set to the remaining available 8th bits, computed by taking the size of the coded frame times 8 and subtracting ec_tell_frac(). From this value, one (8th bit) is subtracted to ensure that the resulting allocation will be conservative. 'anti_collapse_rsv' is set to 8 (8th bits) iff the frame is a transient, LM is greater than 1, and total is greater than or equal to (LM+2) * 8. Total is then decremented by anti_collapse_rsv and clamped to be equal to or greater than zero. 'skip_rsv' is set to 8 (8th bits) if total is greater than 8, otherwise it is zero. Total is then decremented by skip_rsv. This reserves space for the final skipping flag.</t> <t>If the current frame is stereo, intensity_rsv is set to the conservative log2 in 8th bits of the number of coded bands for this frame (given by the table LOG2_FRAC_TABLE). If intensity_rsv is greater than total then intensity_rsv is set to zero. Otherwise total is decremented by intensity_rsv, and if total is still greater than 8, dual_stereo_rsv is set to 8 and total is decremented by dual_stereo_rsv.</t> <t>The allocation process then computes a vector representing the hard minimum amounts allocation any band will receive for shape. This minimum is higher than the technical limit of the PVQ process, but very low rate allocations produce an excessively sparse spectrum and these bands are better served by having no allocation at all. For each coded band, set thresh[band] to twenty-four times the number of MDCT bins in the band and divide by 16. If 8 times the number of channels is greater, use that instead. This sets the minimum allocation to one bit per channel or 48 128th bits per MDCT bin, whichever is greater. The band-size dependent part of this value is not scaled by the channel count, because at the very low rates where this limit is applicable there will usually be no bits allocated to the side.</t> <t>The previously decoded allocation trim is used to derive a vector of per-band adjustments, 'trim_offsets[]'. For each coded band take the alloc_trim and subtract 5 and LM. Then multiply the result by the number of channels, the number of MDCT bins in the shortest frame size for this mode, the number of remaining bands, 2**LM, and 8. Then divide this value by 64. Finally, if the number of MDCT bins in the band per channel is only one, 8 times the number of channels is subtracted in order to diminish the allocation by one bit, because width 1 bands receive greater benefit from the coarse energy coding.</t> </section> <section anchor="PVQ-decoder" title="Shape Decoder"> <t> In each band, the normalized "shape" is encoded using a vector quantization scheme called a "pyramid vector quantizer". </t> <t>In the simplest case, the number of bits allocated in <xref target="allocation"></xref> is converted to a number of pulses as described by <xref target="bits-pulses"></xref>. Knowing the number of pulses and the number of samples in the band, the decoder calculates the size of the codebook as detailed in <xref target="cwrs-decoder"></xref>. The size is used to decode an unsigned integer (uniform probability model), which is the codeword index. This index is converted into the corresponding vector as explained in <xref target="cwrs-decoder"></xref>. This vector is then scaled to unit norm. </t> <section anchor="bits-pulses" title="Bits to Pulses"> <t> Although the allocation is performed in 1/8th bit units, the quantization requires an integer number of pulses K. To do this, the encoder searches for the value of K that produces the number of bits nearest to the allocated value (rounding down if exactly halfway between two values), not to exceed the total number of bits available. For efficiency reasons, the search is performed against a precomputed allocation table which only permits some K values for each N. The number of codebook entries can be computed as explained in <xref target="cwrs-decoder"></xref>. The difference between the number of bits allocated and the number of bits used is accumulated to a "balance" (initialized to zero) that helps adjust the allocation for the next bands. One third of the balance is applied to the bit allocation of each band to help achieve the target allocation. The only exceptions are the band before the last and the last band, for which half the balance and the whole balance are applied, respectively. </t> </section> <section anchor="cwrs-decoder" title="PVQ Decoding"> <t> Decoding of PVQ vectors is implemented in decode_pulses() (cwrs.c). The uique codeword index is decoded as a uniformly-distributed integer value between 0 and V(N,K)-1, where V(N,K) is the number of possible combinations of K pulses in N samples. The index is then converted to a vector in the same way specified in <xref target="PVQ"></xref>. The indexing is based on the calculation of V(N,K) (denoted N(L,K) in <xref target="PVQ"></xref>). </t> <t> The number of combinations can be computed recursively as V(N,K) = V(N-1,K) + V(N,K-1) + V(N-1,K-1), with V(N,0) = 1 and V(0,K) = 0, K != 0. There are many different ways to compute V(N,K), including precomputed tables and direct use of the recursive formulation. The reference implementation applies the recursive formulation one line (or column) at a time to save on memory use, along with an alternate, univariate recurrence to initialize an arbitrary line, and direct polynomial solutions for small N. All of these methods are equivalent, and have different trade-offs in speed, memory usage, and code size. Implementations MAY use any methods they like, as long as they are equivalent to the mathematical definition. </t> <t> The decoded vector is normalised such that its L2-norm equals one. </t> </section> <section anchor="spreading" title="Spreading"> <t> The normalised vector decoded in <xref target="cwrs-decoder"/> is then rotated for the purpose of avoiding tonal artefacts. The rotation gain is equal to <figure align="center"> <artwork align="center"><![CDATA[ g_r = N / (N + f_r*K) ]]></artwork> </figure> where N is the number of dimensions, K is the number of pulses, and f_r depends on the value of the "spread" parameter in the bit-stream. </t> <texttable anchor="spread values" title="Spreading values"> <ttcol>Spread value</ttcol> <ttcol>f_r</ttcol> <c>0</c> <c>infinite (no rotation)</c> <c>1</c> <c>15</c> <c>2</c> <c>10</c> <c>3</c> <c>5</c> </texttable> <t> The rotation angle is then calculated as <figure align="center"> <artwork align="center"><![CDATA[ 2 pi * g_r theta = ---------- 4 ]]></artwork> </figure> A 2-D rotation R(i,j) between points x_i and x_j is defined as: <figure align="center"> <artwork align="center"><![CDATA[ x_i' = cos(theta)*x_i + sin(theta)*x_j x_j' = -sin(theta)*x_i + cos(theta)*x_j ]]></artwork> </figure> An N-D rotation is then achieved by applying a series of 2-D rotations back and forth, in the following order: R(x_1, x_2), R(x_2, x_3), ..., R(x_N-2, X_N-1), R(x_N-1, X_N), R(x_N-2, X_N-1), ..., R(x_1, x_2). </t> <t> If the decoded vector represents more than one time block, then the following process is applied separately on each time block. Also, if each block represents 8 samples or more, then another N-D rotation, by (pi/2-theta), is applied <spanx style="emph">before</spanx> the rotation described above. This extra rotation is applied in an interleaved manner with a stride equal to round(sqrt(N/nb_blocks)) </t> </section> <section anchor="split" title="Split decoding"> <t> To avoid the need for multi-precision calculations when decoding PVQ codevectors, the maximum size allowed for codebooks is 32 bits. When larger codebooks are needed, the vector is instead split in two sub-vectors of size N/2. A quantized gain parameter with precision derived from the current allocation is entropy coded to represent the relative gains of each side of the split, and the entire decoding process is recursively applied. Multiple levels of splitting may be applied up to a frame size dependent limit. The same recursive mechanism is applied for the joint coding of stereo audio. </t> </section> <section anchor="tf-change" title="Time-Frequency change"> <t> The time-frequency (TF) parameters are used to control the time-frequency resolution tradeoff in each coded band. For each band, there are two possible TF choices. For the first band coded, the PDF is {3, 1}/4 for frames marked as transient and {15, 1}/16 for the other frames. For subsequent bands, the TF choice is coded relative to the previous TF choice with probability {15, 1}/15 for transient frames and {31, 1}/32 otherwise. The mapping between the decoded TF choices and the adjustment in TF resolution is shown in the tables below. </t> <texttable anchor='tf_00'> <ttcol align='center'>Frame size (ms)</ttcol> <ttcol align='center'>0</ttcol> <ttcol align='center'>1</ttcol> <c>2.5</c> <c>0</c> <c>-1</c> <c>5</c> <c>0</c> <c>-1</c> <c>10</c> <c>0</c> <c>-2</c> <c>20</c> <c>0</c> <c>-2</c> <postamble>TF adjustments for non-transient frames and tf_select=0</postamble> </texttable> <texttable anchor='tf_01'> <ttcol align='center'>Frame size (ms)</ttcol> <ttcol align='center'>0</ttcol> <ttcol align='center'>1</ttcol> <c>2.5</c> <c>0</c> <c>-1</c> <c>5</c> <c>0</c> <c>-2</c> <c>10</c> <c>0</c> <c>-3</c> <c>20</c> <c>0</c> <c>-3</c> <postamble>TF adjustments for non-transient frames and tf_select=1</postamble> </texttable> <texttable anchor='tf_10'> <ttcol align='center'>Frame size (ms)</ttcol> <ttcol align='center'>0</ttcol> <ttcol align='center'>1</ttcol> <c>2.5</c> <c>0</c> <c>-1</c> <c>5</c> <c>1</c> <c>0</c> <c>10</c> <c>2</c> <c>0</c> <c>20</c> <c>3</c> <c>0</c> <postamble>TF adjustments for transient frames and tf_select=0</postamble> </texttable> <texttable anchor='tf_11'> <ttcol align='center'>Frame size (ms)</ttcol> <ttcol align='center'>0</ttcol> <ttcol align='center'>1</ttcol> <c>2.5</c> <c>0</c> <c>-1</c> <c>5</c> <c>1</c> <c>-1</c> <c>10</c> <c>1</c> <c>-1</c> <c>20</c> <c>1</c> <c>-1</c> <postamble>TF adjustments for transient frames and tf_select=1</postamble> </texttable> <t> A negative TF adjustment means that the temporal resolution is increased, while a positive TF adjustment means that the frequency resolution is increased. Changes in TF resolution are implemented using the Hadamard transform. To increase the time resolution by N, N "levels" of the Hadamard transform are applied to the decoded vector for each interleaved MDCT vector. To increase the frequency resolution (assumes a transient frame), then N levels of the Hadamard transform are applied <spanx style="emph">across</spanx> the interleaved MDCT vector. In the case of increased time resolution the decoder uses the "sequency order" because the input vector is sorted in time. </t> </section> </section> <section anchor="anti-collapse" title="Anti-collapse processing"> <t> When the frame has the transient bit set, an anti-collapse bit is decoded. When anti-collapse is set, the energy in each small MDCT is prevented from collapsing to zero. For each band of each MDCT where a collapse is detected, a pseudo-random signal is inserted with an energy corresponding to the min energy over the two previous frames. A renormalization step is then required to ensure that the anti-collapse step did not alter the energy preservation property. </t> </section> <section anchor="denormalization" title="Denormalization"> <t> Just like each band was normalized in the encoder, the last step of the decoder before the inverse MDCT is to denormalize the bands. Each decoded normalized band is multiplied by the square root of the decoded energy. This is done by denormalise_bands() (bands.c). </t> </section> <section anchor="inverse-mdct" title="Inverse MDCT"> <t>The inverse MDCT implementation has no special characteristics. The input is N frequency-domain samples and the output is 2*N time-domain samples, while scaling by 1/2. A "low-overlap" window is used to reduce the algorithmic delay. It is derived from a basic (full overlap) 240-sample version of the window used by the Vorbis codec: <figure align="center"> <artwork align="center"><![CDATA[ 2 / /pi /pi n + 1/2\ \ \ W(n) = |sin|-- * sin|-- * -------| | | . \ \2 \2 L / / / ]]></artwork> </figure> The low-overlap window is created by zero-padding the basic window and inserting ones in the middle, such that the resulting window still satisfies power complementarity. The IMDCT and windowing are performed by mdct_backward (mdct.c). </t> <section anchor="post-filter" title="Post-filter"> <t> The output of the inverse MDCT (after weighted overlap-add) is sent to the post-filter. Although the post-filter is applied at the end, the post-filter parameters are encoded at the beginning, just after the silence flag. The post-filter can be switched on or off using one bit (logp=1). If the post-filter is enabled, then the octave is decoded as an integer value between 0 and 6 of uniform probability. Once the octave is known, the fine pitch within the octave is decoded using 4+octave raw bits. The final pitch period is equal to (16<<octave)+fine_pitch-1 so it is bounded between 15 and 1022, inclusively. Next, the gain is decoded as three raw bits and is equal to G=3*(int_gain+1)/32. The set of post-filter taps is decoded last, using a pdf equal to {2, 1, 1}/4. Tapset zero corresponds to the filter coefficients g0 = 0.3066406250, g1 = 0.2170410156, g2 = 0.1296386719. Tapset one corresponds to the filter coefficients g0 = 0.4638671875, g1 = 0.2680664062, g2 = 0, and tapset two uses filter coefficients g0 = 0.7998046875, g1 = 0.1000976562, g2 = 0. </t> <t> The post-filter response is thus computed as: <figure align="center"> <artwork align="center"> <![CDATA[ y(n) = x(n) + G*(g0*y(n-T) + g1*(y(n-T+1)+y(n-T+1)) + g2*(y(n-T+2)+y(n-T+2))) ]]> </artwork> </figure> During a transition between different gains, a smooth transition is calculated using the square of the MDCT window. It is important that values of y(n) be interpolated one at a time such that the past value of y(n) used is interpolated. </t> </section> <section anchor="deemphasis" title="De-emphasis"> <t> After the post-filter, the signal is de-emphasized using the inverse of the pre-emphasis filter used in the encoder: <figure align="center"> <artwork align="center"><![CDATA[ 1 1 ---- = --------------- , A(z) -1 1 - alpha_p*z ]]></artwork> </figure> where alpha_p=0.8500061035. </t> </section> </section> </section> <section anchor="Packet Loss Concealment" title="Packet Loss Concealment (PLC)"> <t> Packet loss concealment (PLC) is an optional decoder-side feature that SHOULD be included when receiving from an unreliable channel. Because PLC is not part of the bitstream, there are many acceptable ways to implement PLC with different complexity/quality trade-offs. </t> <t> The PLC in the reference implementation depends on the mode of last packet received. In CELT mode, the PLC finds a periodicity in the decoded signal and repeats the windowed waveform using the pitch offset. The windowed waveform is overlapped in such a way as to preserve the time-domain aliasing cancellation with the previous frame and the next frame. This is implemented in celt_decode_lost() (mdct.c). In SILK mode, the PLC uses LPC extrapolation from the previous frame, implemented in silk_PLC() (PLC.c). </t> <section anchor="clock-drift" title="Clock Drift Compensation"> <t> Clock drift refers to the gradual desynchronization of two endpoints whose sample clocks run at different frequencies while they are streaming live audio. Differences in clock frequencies are generally attributable to manufacturing variation in the endpoints' clock hardware. For long-lived streams, the time difference between sender and receiver can grow without bound. </t> <t> When the sender's clock runs slower than the receiver's, the effect is similar to packet loss: too few packets are received. The receiver can distinguish between drift and loss if the transport provides packet timestamps. A receiver for live streams SHOULD conceal the effects of drift, and MAY do so by invoking the PLC. </t> <t> When the sender's clock runs faster than the receiver's, too many packets will be received. The receiver MAY respond by skipping any packet (i.e. not submitting the packet for decoding). This is likely to produce a less severe artifact than if the frame were dropped after decoding. </t> <t> A decoder MAY employ a more sophisticated drift compensation method. For example, the <eref target='http://code.google.com/p/webrtc/source/browse/trunk/src/modules/audio_coding/NetEQ/main/source/?r=583'>NetEQ component</eref> of the <eref target='http://code.google.com/p/webrtc/'>WebRTC.org codebase</eref> compensates for drift by adding or removing one period when the signal is highly periodic. The reference implementation of Opus allows a caller to learn whether the current frame's signal is highly periodic, and if so what the period is, using the OPUS_GET_PITCH() request. </t> </section> </section> <section anchor="switching" title="Mode Switching"> <t> Switching between the Opus coding modes requires careful consideration. More specifically, the transitions that cannot be easily handled are the ones where the lower frequencies have to switch between the SILK LP-based model and the CELT transform model. If nothing is done, a glitch will occur for these transitions. On the other hand, switching between the SILK-only modes and the hybrid mode does not require any special treatment. </t> <t> There are two ways to avoid or reduce glitches during the problematic mode transitions: with side information or without it. Only transitions with side information are normatively specified. For transitions with no side information, it is RECOMMENDED for the decoder to use a concealment technique (e.g. make use of the PLC algorithm) to "fill in" the gap or discontinuity caused by the mode transition. Note that this concealment MUST NOT be applied when switching between the SILK mode and the hybrid mode or vice versa. Similarly, it MUST NOT be applied when merely changing the bandwidth within the same mode. </t> <section anchor="side-info" title="Switching Side Information"> <t> Switching with side information involves transmitting in-band a 5-ms "redundant" CELT frame within the Opus frame. This frame is designed to fill in the gap or discontinuity without requiring the decoder to conceal it. For transitions from a CELT-only frame to a SILK-only or hybrid frame, the redundant frame is inserted in the frame following the transition (i.e. the SILK-only/hybrid frame). For transitions from a SILK-only/hybrid frame to a CELT-only frame, the redundant frame is inserted in the first frame. For all SILK-only and hybrid frames (not only those involved in a mode transition), a binary symbol of probability 2^-12 needs to be decoded just after the SILK part of the bitstream. When the symbol value is 1, the frame then includes an embedded redundant frame. The redundant frame always starts and ends on a byte boundary. For SILK-only frames, the number of bytes is simply the number of whole remaining bytes. For hybrid frames, the number of bytes is equal to 2, plus a decoded unsigned integer (ec_dec_uint()) between 0 and 255. For hybrid frames, the redundant frame is placed at the end of the frame, after the CELT layer of the hybrid frame. The redundant frame is decoded like any other CELT-only frame, with the exception that it does not contain a TOC byte. The bandwidth is instead set to the same bandwidth of the current frame (for MB frames, the redundant frame is set to WB). </t> <t> For CELT-only to SILK-only/hybrid transitions, the first 2.5 ms of the redundant frame is used as-is for the reconstructed output. The remaining 2.5 ms is overlapped and added (cross-faded using the square of the MDCT power-complementary window) to the decoded SILK/hybrid signal, ensuring a smooth transition. For SILK-only/hyrid to CELT-only transitions, only the second half of the 5-ms decoded redundant frame is used. In that case, only a 2.5-ms cross-fade is applied, still using the power-complementary window. </t> </section> </section> </section> <!-- ******************************************************************* --> <!-- ************************** OPUS ENCODER *********************** --> <!-- ******************************************************************* --> <section title="Codec Encoder"> <t> Opus encoder block diagram. <figure> <artwork> <![CDATA[ +----------+ +-------+ | sample | | SILK | +->| rate |--->|encoder|--+ | |conversion| | | | audio | +----------+ +-------+ | +-------+ ------+ +--->| Range | | +------------+ +-------+ |encoder|----> | | Delay | | CELT | +--->| | bitstream +->|Compensation|->|encoder|--+ +-------+ | | | | +------------+ +-------+ ]]> </artwork> </figure> </t> <section anchor="range-encoder" title="Range Coder"> <t> The range coder also acts as the bit-packer for Opus. It is used in three different ways, to encode: <list style="symbols"> <t>entropy-coded symbols with a fixed probability model using ec_encode(), (entenc.c)</t> <t>integers from 0 to 2**M-1 using ec_enc_uint() or ec_enc_bits(), (entenc.c)</t> <t>integers from 0 to N-1 (where N is not a power of two) using ec_enc_uint(). (entenc.c)</t> </list> </t> <t> The range encoder maintains an internal state vector composed of the four-tuple (low,rng,rem,ext) representing the low end of the current range, the size of the current range, a single buffered output octet, and a count of additional carry-propagating output octets. Both rng and low are 32-bit unsigned integer values, rem is an octet value or the special value -1, and ext is an integer with at least 16 bits. This state vector is initialized at the start of each each frame to the value (0,2**31,-1,0). The reference implementation re-uses the 'val' field of the entropy coder structure to hold low, in order to allow the same structure to be used for encoding and decoding, but we maintain the distinction here for clarity. </t> <section anchor="encoding-symbols" title="Encoding Symbols"> <t> The main encoding function is ec_encode() (entenc.c), which takes as an argument a three-tuple (fl,fh,ft) describing the range of the symbol to be encoded in the current context, with 0 <= fl < fh <= ft <= 65535. The values of this tuple are derived from the probability model for the symbol. Let f(i) be the frequency of the i'th symbol in the current context. Then the three-tuple corresponding to the k'th symbol is given by <![CDATA[ fl=sum(f(i),i<k), fh=fl+f(i), and ft=sum(f(i)). ]]> </t> <t> ec_encode() updates the state of the encoder as follows. If fl is greater than zero, then low = low + rng - (rng/ft)*(ft-fl) and rng = (rng/ft)*(fh-fl). Otherwise, low is unchanged and rng = rng - (rng/ft)*(fh-fl). The divisions here are exact integer division. After this update, the range is normalized. </t> <t> To normalize the range, the following process is repeated until rng > 2**23. First, the top 9 bits of low, (low>>23), are placed into a carry buffer. Then, low is set to <![CDATA[(low << 8 & 0x7FFFFFFF) and rng is set to (rng<<8)]]>. This process is carried out by ec_enc_normalize() (entenc.c). </t> <t> The 9 bits produced in each iteration of the normalization loop consist of 8 data bits and a carry flag. The final value of the output bits is not determined until carry propagation is accounted for. Therefore the reference implementation buffers a single (non-propagating) output octet and keeps a count of additional propagating (0xFF) output octets. An implementation MAY choose to use any mathematically equivalent scheme to perform carry propagation. </t> <t> The function ec_enc_carry_out() (entenc.c) performs this buffering. It takes a 9-bit input value, c, from the normalization: 8 bits of output and a carry bit. If c is 0xFF, then ext is incremented and no octets are output. Otherwise, if rem is not the special value -1, then the octet (rem+(c>>8)) is output. Then ext octets are output with the value 0 if the carry bit is set, or 0xFF if it is not, and rem is set to the lower 8 bits of c. After this, ext is set to zero. </t> <t> In the reference implementation, a special version of ec_encode() called ec_encode_bin() (entenc.c) is defined to take a two-tuple (fl,ftb), where <![CDATA[0 <= fl < 2**ftb and ftb < 16. It is mathematically equivalent to calling ec_encode() with the three-tuple (fl,fl+1,1<<ftb)]]>, but avoids using division. </t> </section> <section anchor="encoding-bits" title="Encoding Raw Bits"> <t> The CELT layer also allows directly encoding a series of raw bits, outside of the range coder, implemented in ec_enc_bits() (entenc.c). The raw bits are packed at the end of the packet, starting by storing the least significant bit of the value to be packed in the least significant bit of the last byte, filling up to the most significant bit in the last byte, and then continuing in the least significant bit of the penultimate byte, and so on. This packing may continue into the last byte output by the range coder, though the format should render it impossible to overwrite any set bit produced by the range coder when the procedure in <xref target='encoder-finalizing'/> is followed to finalize the stream. </t> </section> <section anchor="encoding-ints" title="Encoding Uniformly Distributed Integers"> <t> The function ec_enc_uint() is based on ec_encode() and encodes one of N equiprobable symbols, each with a frequency of 1, where N may be as large as 2**32-1. Because ec_encode() is limited to a total frequency of 2**16-1, this is done by encoding a series of symbols in smaller contexts. </t> <t> ec_enc_uint() (entenc.c) takes a two-tuple (fl,ft), where ft is not necessarily a power of two. Let ftb be the location of the highest 1 bit in the two's-complement representation of (ft-1), or -1 if no bits are set. If ftb>8, then the top 8 bits of fl are encoded using ec_encode() with the three-tuple (fl>>ftb-8,(fl>>ftb-8)+1,(ft-1>>ftb-8)+1), and the remaining bits are encoded as raw bits. Otherwise, fl is encoded with ec_encode() directly using the three-tuple (fl,fl+1,ft). </t> </section> <section anchor="encoder-finalizing" title="Finalizing the Stream"> <t> After all symbols are encoded, the stream must be finalized by outputting a value inside the current range. Let end be the integer in the interval [low,low+rng) with the largest number of trailing zero bits, b, such that end+(1<<b)-1 is also in the interval [low,low+rng). Then while end is not zero, the top 9 bits of end, e.g., <![CDATA[(end>>23), are sent to the carry buffer, and end is replaced by (end<<8&0x7FFFFFFF). Finally, if the value in carry buffer, rem, is]]> neither zero nor the special value -1, or the carry count, ext, is greater than zero, then 9 zero bits are sent to the carry buffer. After the carry buffer is finished outputting octets, the rest of the output buffer (if any) is padded with zero bits, until it reaches the raw bits. Finally, rem is set to the special value -1. This process is implemented by ec_enc_done() (entenc.c). </t> </section> <section anchor="encoder-tell" title="Current Bit Usage"> <t> The bit allocation routines in Opus need to be able to determine a conservative upper bound on the number of bits that have been used to encode the current frame thus far. This drives allocation decisions and ensures that the range coder and raw bits will not overflow the output buffer. This is computed in the reference implementation to whole-bit precision by the function ec_tell() (entcode.h) and to fractional 1/8th bit precision by the function ec_tell_frac() (entcode.c). Like all operations in the range coder, it must be implemented in a bit-exact manner, and must produce exactly the same value returned by the same functions in the decoder after decoding the same symbols. </t> </section> </section> <section title='SILK Encoder'> <t> In the following, we focus on the core encoder and describe its components. For simplicity, we will refer to the core encoder simply as the encoder in the remainder of this document. An overview of the encoder is given in <xref target="encoder_figure" />. </t> <figure align="center" anchor="encoder_figure"> <artwork align="center"> <![CDATA[ +---+ +----------------------------->| | +---------+ | +---------+ | | |Voice | | |LTP | | | +----->|Activity |-----+ +---->|Scaling |---------+--->| | | |Detector | 3 | | |Control |<+ 12 | | | | +---------+ | | +---------+ | | | | | | | +---------+ | | | | | | | |Gains | | 11 | | | | | | +->|Processor|-|---+---|--->| R | | | | | | | | | | | a | | \/ | | +---------+ | | | | n | | +---------+ | | +---------+ | | | | g | | |Pitch | | | |LSF | | | | | e | | +->|Analysis |-+ | |Quantizer|-|---|---|--->| | | | | |4| | | | | 8 | | | E |-> | | +---------+ | | +---------+ | | | | n |14 | | | | 9/\ 10| | | | | c | | | | | | \/ | | | | o | | | +---------+ | | +----------+| | | | d | | | |Noise | +--|->|Prediction|+---|---|--->| e | | +->|Shaping |-|--+ |Analysis || 7 | | | r | | | |Analysis |5| | | || | | | | | | +---------+ | | +----------+| | | | | | | | | /\ | | | | | | | +---------|--|-------+ | | | | | | | | \/ \/ \/ \/ \/ | | | +---------+ | | +---------+ +------------+ | | | |High-Pass| | | | | |Noise | | | -+->|Filter |-+----+----->|Prefilter|------>|Shaping |->| | 1 | | 2 | | 6 |Quantization|13| | +---------+ +---------+ +------------+ +---+ 1: Input speech signal 2: High passed input signal 3: Voice activity estimate 4: Pitch lags (per 5 ms) and voicing decision (per 20 ms) 5: Noise shaping quantization coefficients - Short term synthesis and analysis noise shaping coefficients (per 5 ms) - Long term synthesis and analysis noise shaping coefficients (per 5 ms and for voiced speech only) - Noise shaping tilt (per 5 ms) - Quantizer gain/step size (per 5 ms) 6: Input signal filtered with analysis noise shaping filters 7: Short and long term prediction coefficients LTP (per 5 ms) and LPC (per 20 ms) 8: LSF quantization indices 9: LSF coefficients 10: Quantized LSF coefficients 11: Processed gains, and synthesis noise shape coefficients 12: LTP state scaling coefficient. Controlling error propagation / prediction gain trade-off 13: Quantized signal 14: Range encoded bitstream ]]> </artwork> <postamble>Encoder block diagram.</postamble> </figure> <section title='Voice Activity Detection'> <t> The input signal is processed by a Voice Activity Detector (VAD) to produce a measure of voice activity, spectral tilt, and signal-to-noise estimates for each frame. The VAD uses a sequence of half-band filterbanks to split the signal into four subbands: 0 - Fs/16, Fs/16 - Fs/8, Fs/8 - Fs/4, and Fs/4 - Fs/2, where Fs is the sampling frequency (8, 12, 16, or 24 kHz). The lowest subband, from 0 - Fs/16, is high-pass filtered with a first-order moving average (MA) filter (with transfer function H(z) = 1-z**(-1)) to reduce the energy at the lowest frequencies. For each frame, the signal energy per subband is computed. In each subband, a noise level estimator tracks the background noise level and a Signal-to-Noise Ratio (SNR) value is computed as the logarithm of the ratio of energy to noise level. Using these intermediate variables, the following parameters are calculated for use in other SILK modules: <list style="symbols"> <t> Average SNR. The average of the subband SNR values. </t> <t> Smoothed subband SNRs. Temporally smoothed subband SNR values. </t> <t> Speech activity level. Based on the average SNR and a weighted average of the subband energies. </t> <t> Spectral tilt. A weighted average of the subband SNRs, with positive weights for the low subbands and negative weights for the high subbands. </t> </list> </t> </section> <section title='High-Pass Filter'> <t> The input signal is filtered by a high-pass filter to remove the lowest part of the spectrum that contains little speech energy and may contain background noise. This is a second order Auto Regressive Moving Average (ARMA) filter with a cut-off frequency around 70 Hz. </t> <t> In the future, a music detector may also be used to lower the cut-off frequency when the input signal is detected to be music rather than speech. </t> </section> <section title='Pitch Analysis' anchor='pitch_estimator_overview_section'> <t> The high-passed input signal is processed by the open loop pitch estimator shown in <xref target='pitch_estimator_figure' />. <figure align="center" anchor="pitch_estimator_figure"> <artwork align="center"> <![CDATA[ +--------+ +----------+ |2 x Down| |Time- | +->|sampling|->|Correlator| | | | | | | |4 | +--------+ +----------+ \/ | | 2 +-------+ | | +-->|Speech |5 +---------+ +--------+ | \/ | |Type |-> |LPC | |Down | | +----------+ | | +->|Analysis | +->|sample |-+------------->|Time- | +-------+ | | | | |to 8 kHz| |Correlator|-----------> | +---------+ | +--------+ |__________| 6 | | | |3 | \/ | \/ | +---------+ | +----------+ | |Whitening| | |Time- | -+->|Filter |-+--------------------------->|Correlator|-----------> 1 | | | | 7 +---------+ +----------+ 1: Input signal 2: Lag candidates from stage 1 3: Lag candidates from stage 2 4: Correlation threshold 5: Voiced/unvoiced flag 6: Pitch correlation 7: Pitch lags ]]> </artwork> <postamble>Block diagram of the pitch estimator.</postamble> </figure> The pitch analysis finds a binary voiced/unvoiced classification, and, for frames classified as voiced, four pitch lags per frame - one for each 5 ms subframe - and a pitch correlation indicating the periodicity of the signal. The input is first whitened using a Linear Prediction (LP) whitening filter, where the coefficients are computed through standard Linear Prediction Coding (LPC) analysis. The order of the whitening filter is 16 for best results, but is reduced to 12 for medium complexity and 8 for low complexity modes. The whitened signal is analyzed to find pitch lags for which the time correlation is high. The analysis consists of three stages for reducing the complexity: <list style="symbols"> <t>In the first stage, the whitened signal is downsampled to 4 kHz (from 8 kHz) and the current frame is correlated to a signal delayed by a range of lags, starting from a shortest lag corresponding to 500 Hz, to a longest lag corresponding to 56 Hz.</t> <t> The second stage operates on an 8 kHz signal (downsampled from 12, 16, or 24 kHz) and measures time correlations only near the lags corresponding to those that had sufficiently high correlations in the first stage. The resulting correlations are adjusted for a small bias towards short lags to avoid ending up with a multiple of the true pitch lag. The highest adjusted correlation is compared to a threshold depending on: <list style="symbols"> <t> Whether the previous frame was classified as voiced </t> <t> The speech activity level </t> <t> The spectral tilt. </t> </list> If the threshold is exceeded, the current frame is classified as voiced and the lag with the highest adjusted correlation is stored for a final pitch analysis of the highest precision in the third stage. </t> <t> The last stage operates directly on the whitened input signal to compute time correlations for each of the four subframes independently in a narrow range around the lag with highest correlation from the second stage. </t> </list> </t> </section> <section title='Noise Shaping Analysis' anchor='noise_shaping_analysis_overview_section'> <t> The noise shaping analysis finds gains and filter coefficients used in the prefilter and noise shaping quantizer. These parameters are chosen such that they will fulfill several requirements: <list style="symbols"> <t>Balancing quantization noise and bitrate. The quantization gains determine the step size between reconstruction levels of the excitation signal. Therefore, increasing the quantization gain amplifies quantization noise, but also reduces the bitrate by lowering the entropy of the quantization indices.</t> <t>Spectral shaping of the quantization noise; the noise shaping quantizer is capable of reducing quantization noise in some parts of the spectrum at the cost of increased noise in other parts without substantially changing the bitrate. By shaping the noise such that it follows the signal spectrum, it becomes less audible. In practice, best results are obtained by making the shape of the noise spectrum slightly flatter than the signal spectrum.</t> <t>De-emphasizing spectral valleys; by using different coefficients in the analysis and synthesis part of the prefilter and noise shaping quantizer, the levels of the spectral valleys can be decreased relative to the levels of the spectral peaks such as speech formants and harmonics. This reduces the entropy of the signal, which is the difference between the coded signal and the quantization noise, thus lowering the bitrate.</t> <t>Matching the levels of the decoded speech formants to the levels of the original speech formants; an adjustment gain and a first order tilt coefficient are computed to compensate for the effect of the noise shaping quantization on the level and spectral tilt.</t> </list> </t> <t> <figure align="center" anchor="noise_shape_analysis_spectra_figure"> <artwork align="center"> <![CDATA[ / \ ___ | // \\ | // \\ ____ |_// \\___// \\ ____ | / ___ \ / \\ // \\ P |/ / \ \_/ \\_____// \\ o | / \ ____ \ / \\ w | / \___/ \ \___/ ____ \\___ 1 e |/ \ / \ \ r | \_____/ \ \__ 2 | \ | \___ 3 | +----------------------------------------> Frequency 1: Input signal spectrum 2: De-emphasized and level matched spectrum 3: Quantization noise spectrum ]]> </artwork> <postamble>Noise shaping and spectral de-emphasis illustration.</postamble> </figure> <xref target='noise_shape_analysis_spectra_figure' /> shows an example of an input signal spectrum (1). After de-emphasis and level matching, the spectrum has deeper valleys (2). The quantization noise spectrum (3) more or less follows the input signal spectrum, while having slightly less pronounced peaks. The entropy, which provides a lower bound on the bitrate for encoding the excitation signal, is proportional to the area between the de-emphasized spectrum (2) and the quantization noise spectrum (3). Without de-emphasis, the entropy is proportional to the area between input spectrum (1) and quantization noise (3) - clearly higher. </t> <t> The transformation from input signal to de-emphasized signal can be described as a filtering operation with a filter <figure align="center"> <artwork align="center"> <![CDATA[ -1 Wana(z) H(z) = G * ( 1 - c_tilt * z ) * ------- Wsyn(z), ]]> </artwork> </figure> having an adjustment gain G, a first order tilt adjustment filter with tilt coefficient c_tilt, and where <figure align="center"> <artwork align="center"> <![CDATA[ 16 d __ -k -L __ -k Wana(z) = (1 - \ (a_ana(k) * z )*(1 - z * \ b_ana(k) * z ), /_ /_ k=1 k=-d ]]> </artwork> </figure> is the analysis part of the de-emphasis filter, consisting of the short-term shaping filter with coefficients a_ana(k), and the long-term shaping filter with coefficients b_ana(k) and pitch lag L. The parameter d determines the number of long-term shaping filter taps. </t> <t> Similarly, but without the tilt adjustment, the synthesis part can be written as <figure align="center"> <artwork align="center"> <![CDATA[ 16 d __ -k -L __ -k Wsyn(z) = (1 - \ (a_syn(k) * z )*(1 - z * \ b_syn(k) * z ). /_ /_ k=1 k=-d ]]> </artwork> </figure> </t> <t> All noise shaping parameters are computed and applied per subframe of 5 ms. First, an LPC analysis is performed on a windowed signal block of 15 ms. The signal block has a look-ahead of 5 ms relative to the current subframe, and the window is an asymmetric sine window. The LPC analysis is done with the autocorrelation method, with an order of 16 for best quality or 12 in low complexity operation. The quantization gain is found by taking the square root of the residual energy from the LPC analysis and multiplying it by a value inversely proportional to the coding quality control parameter and the pitch correlation. </t> <t> Next we find the two sets of short-term noise shaping coefficients a_ana(k) and a_syn(k), by applying different amounts of bandwidth expansion to the coefficients found in the LPC analysis. This bandwidth expansion moves the roots of the LPC polynomial towards the origin, using the formulas <figure align="center"> <artwork align="center"> <![CDATA[ k a_ana(k) = a(k)*g_ana , and k a_syn(k) = a(k)*g_syn , ]]> </artwork> </figure> where a(k) is the k'th LPC coefficient, and the bandwidth expansion factors g_ana and g_syn are calculated as <figure align="center"> <artwork align="center"> <![CDATA[ g_ana = 0.94 - 0.02*C, and g_syn = 0.94 + 0.02*C, ]]> </artwork> </figure> where C is the coding quality control parameter between 0 and 1. Applying more bandwidth expansion to the analysis part than to the synthesis part gives the desired de-emphasis of spectral valleys in between formants. </t> <t> The long-term shaping is applied only during voiced frames. It uses three filter taps, described by <figure align="center"> <artwork align="center"> <![CDATA[ b_ana = F_ana * [0.25, 0.5, 0.25], and b_syn = F_syn * [0.25, 0.5, 0.25]. ]]> </artwork> </figure> For unvoiced frames these coefficients are set to 0. The multiplication factors F_ana and F_syn are chosen between 0 and 1, depending on the coding quality control parameter, as well as the calculated pitch correlation and smoothed subband SNR of the lowest subband. By having F_ana less than F_syn, the pitch harmonics are emphasized relative to the valleys in between the harmonics. </t> <t> The tilt coefficient c_tilt is for unvoiced frames chosen as <figure align="center"> <artwork align="center"> <![CDATA[ c_tilt = 0.4, and as c_tilt = 0.04 + 0.06 * C ]]> </artwork> </figure> for voiced frames, where C again is the coding quality control parameter and is between 0 and 1. </t> <t> The adjustment gain G serves to correct any level mismatch between the original and decoded signals that might arise from the noise shaping and de-emphasis. This gain is computed as the ratio of the prediction gain of the short-term analysis and synthesis filter coefficients. The prediction gain of an LPC synthesis filter is the square root of the output energy when the filter is excited by a unit-energy impulse on the input. An efficient way to compute the prediction gain is by first computing the reflection coefficients from the LPC coefficients through the step-down algorithm, and extracting the prediction gain from the reflection coefficients as <figure align="center"> <artwork align="center"> <![CDATA[ K ___ 2 -0.5 predGain = ( | | 1 - (r_k) ) , k=1 ]]> </artwork> </figure> where r_k is the k'th reflection coefficient. </t> <t> Initial values for the quantization gains are computed as the square-root of the residual energy of the LPC analysis, adjusted by the coding quality control parameter. These quantization gains are later adjusted based on the results of the prediction analysis. </t> </section> <section title='Prefilter'> <t> In the prefilter the input signal is filtered using the spectral valley de-emphasis filter coefficients from the noise shaping analysis (see <xref target='noise_shaping_analysis_overview_section'/>). By applying only the noise shaping analysis filter to the input signal, it provides the input to the noise shaping quantizer. </t> </section> <section title='Prediction Analysis' anchor='pred_ana_overview_section'> <t> The prediction analysis is performed in one of two ways depending on how the pitch estimator classified the frame. The processing for voiced and unvoiced speech is described in <xref target='pred_ana_voiced_overview_section' /> and <xref target='pred_ana_unvoiced_overview_section' />, respectively. Inputs to this function include the pre-whitened signal from the pitch estimator (see <xref target='pitch_estimator_overview_section'/>). </t> <section title='Voiced Speech' anchor='pred_ana_voiced_overview_section'> <t> For a frame of voiced speech the pitch pulses will remain dominant in the pre-whitened input signal. Further whitening is desirable as it leads to higher quality at the same available bitrate. To achieve this, a Long-Term Prediction (LTP) analysis is carried out to estimate the coefficients of a fifth-order LTP filter for each of four subframes. The LTP coefficients are used to find an LTP residual signal with the simulated output signal as input to obtain better modeling of the output signal. This LTP residual signal is the input to an LPC analysis where the LPCs are estimated using Burg's method, such that the residual energy is minimized. The estimated LPCs are converted to a Line Spectral Frequency (LSF) vector and quantized as described in <xref target='lsf_quantizer_overview_section' />. After quantization, the quantized LSF vector is converted back to LPC coefficients using the full procedure in <xref target="silk_nlsfs"/>. By using LPC coefficients derived from the quantized LSF coefficients, the encoder remains fully synchronized with the decoder. The LTP coefficients are quantized using a method described in <xref target='ltp_quantizer_overview_section' />. The quantized LPC and LTP coefficients are then used to filter the high-pass filtered input signal and measure residual energy for each of the four subframes. </t> </section> <section title='Unvoiced Speech' anchor='pred_ana_unvoiced_overview_section'> <t> For a speech signal that has been classified as unvoiced, there is no need for LTP filtering, as it has already been determined that the pre-whitened input signal is not periodic enough within the allowed pitch period range for LTP analysis to be worth the cost in terms of complexity and rate. The pre-whitened input signal is therefore discarded, and instead the high-pass filtered input signal is used for LPC analysis using Burg's method. The resulting LPC coefficients are converted to an LSF vector and quantized as described in the following section. They are then transformed back to obtain quantized LPC coefficients, which are then used to filter the high-pass filtered input signal and measure residual energy for each of the four subframes. </t> </section> </section> <section title='LSF Quantization' anchor='lsf_quantizer_overview_section'> <t>In general, the purpose of quantization is to significantly lower the bitrate at the cost of introducing some distortion. A higher rate should always result in lower distortion, and lowering the rate will generally lead to higher distortion. A commonly used but generally suboptimal approach is to use a quantization method with a constant rate, where only the error is minimized when quantizing.</t> <section title='Rate-Distortion Optimization'> <t>Instead, we minimize an objective function that consists of a weighted sum of rate and distortion, and use a codebook with an associated non-uniform rate table. Thus, we take into account that the probability mass function for selecting the codebook entries is by no means guaranteed to be uniform in our scenario. This approach has several advantages. It ensures that rarely used codebook vector centroids, which are modeling statistical outliers in the training set, are quantized with low error at the expense of a high rate. At the same time, it allows modeling frequently used centroids with low error and a relatively low rate. This approach leads to equal or lower distortion than the fixed-rate codebook at any given average rate, provided that the data is similar to that used for training the codebook.</t> </section> <section title='Error Mapping' anchor='lsf_error_mapping_overview_section'> <t> Instead of minimizing the error in the LSF domain, we map the errors to better approximate spectral distortion by applying an individual weight to each element in the error vector. The weight vectors are calculated for each input vector using the Inverse Harmonic Mean Weighting (IHMW) function proposed by Laroia et al. (see <xref target="laroia-icassp" />). Consequently, we solve the following minimization problem, i.e., <figure align="center"> <artwork align="center"> <![CDATA[ LSF_q = argmin { (LSF - c)' * W * (LSF - c) + mu * rate }, c in C ]]> </artwork> </figure> where LSF_q is the quantized vector, LSF is the input vector to be quantized, and c is the quantized LSF vector candidate taken from the set C of all possible outcomes of the codebook. </t> </section> <section title='Survivor Based Codebook Search'> <t> This number of possible combinations is far too high to carry out a full search for each frame, so for all stages but the last (i.e., s smaller than S), only the best min(L, Ms) centroids are carried over to stage s+1. In each stage, the objective function (i.e., the weighted sum of accumulated bitrate and distortion) is evaluated for each codebook vector entry and the results are sorted. Only the best paths and their corresponding quantization errors are considered in the next stage. In the last stage, S, the single best path through the multistage codebook is determined. By varying the maximum number of survivors from each stage to the next, L, the complexity can be adjusted in real time, at the cost of a potential increase when evaluating the objective function for the resulting quantized vector. This approach scales all the way between the two extremes, L=1 being a greedy search, and the desirable but infeasible full search, L=T/MS. Performance almost as good as that of the infeasible full search can be obtained at substantially lower complexity by using this approach (see, e.g., <xref target='leblanc-tsap'/>). </t> </section> <section title='LSF Stabilization' anchor='lsf_stabilizer_overview_section'> <t>If the input is stable, finding the best candidate usually results in a quantized vector that is also stable. Due to the multi-stage approach, however, it is theoretically possible that the best quantization candidate is unstable. Because of this, it is necessary to explicitly ensure that the quantized vectors are stable. Therefore we apply an LSF stabilization method which ensures that the LSF parameters are within valid range, increasingly sorted, and have minimum distances between each other and the border values that have been predetermined as the 0.01 percentile distance values from a large training set.</t> </section> <section title='Off-Line Codebook Training'> <t> The vectors and rate tables for the multi-stage codebook have been trained by minimizing the average of the objective function for LSF vectors from a large training set. </t> </section> </section> <section title='LTP Quantization' anchor='ltp_quantizer_overview_section'> <t> For voiced frames, the prediction analysis described in <xref target='pred_ana_voiced_overview_section' /> resulted in four sets (one set per subframe) of five LTP coefficients, plus four weighting matrices. The LTP coefficients for each subframe are quantized using entropy constrained vector quantization. A total of three vector codebooks are available for quantization, with different rate-distortion trade-offs. The three codebooks have 10, 20, and 40 vectors and average rates of about 3, 4, and 5 bits per vector, respectively. Consequently, the first codebook has larger average quantization distortion at a lower rate, whereas the last codebook has smaller average quantization distortion at a higher rate. Given the weighting matrix W_ltp and LTP vector b, the weighted rate-distortion measure for a codebook vector cb_i with rate r_i is give by <figure align="center"> <artwork align="center"> <![CDATA[ RD = u * (b - cb_i)' * W_ltp * (b - cb_i) + r_i, ]]> </artwork> </figure> where u is a fixed, heuristically-determined parameter balancing the distortion and rate. Which codebook gives the best performance for a given LTP vector depends on the weighting matrix for that LTP vector. For example, for a low valued W_ltp, it is advantageous to use the codebook with 10 vectors as it has a lower average rate. For a large W_ltp, on the other hand, it is often better to use the codebook with 40 vectors, as it is more likely to contain the best codebook vector. The weighting matrix W_ltp depends mostly on two aspects of the input signal. The first is the periodicity of the signal; the more periodic, the larger W_ltp. The second is the change in signal energy in the current subframe, relative to the signal one pitch lag earlier. A decaying energy leads to a larger W_ltp than an increasing energy. Both aspects fluctuate relatively slowly, which causes the W_ltp matrices for different subframes of one frame often to be similar. Because of this, one of the three codebooks typically gives good performance for all subframes, and therefore the codebook search for the subframe LTP vectors is constrained to only allow codebook vectors to be chosen from the same codebook, resulting in a rate reduction. </t> <t> To find the best codebook, each of the three vector codebooks is used to quantize all subframe LTP vectors and produce a combined weighted rate-distortion measure for each vector codebook. The vector codebook with the lowest combined rate-distortion over all subframes is chosen. The quantized LTP vectors are used in the noise shaping quantizer, and the index of the codebook plus the four indices for the four subframe codebook vectors are passed on to the range encoder. </t> </section> <section title='Noise Shaping Quantizer'> <t> The noise shaping quantizer independently shapes the signal and coding noise spectra to obtain a perceptually higher quality at the same bitrate. </t> <t> The prefilter output signal is multiplied with a compensation gain G computed in the noise shaping analysis. Then the output of a synthesis shaping filter is added, and the output of a prediction filter is subtracted to create a residual signal. The residual signal is multiplied by the inverse quantized quantization gain from the noise shaping analysis, and input to a scalar quantizer. The quantization indices of the scalar quantizer represent a signal of pulses that is input to the pyramid range encoder. The scalar quantizer also outputs a quantization signal, which is multiplied by the quantized quantization gain from the noise shaping analysis to create an excitation signal. The output of the prediction filter is added to the excitation signal to form the quantized output signal y(n). The quantized output signal y(n) is input to the synthesis shaping and prediction filters. </t> </section> </section> <section title="CELT Encoder"> <t> Most of the aspects of the CELT encoder can be directly derived from the description of the decoder. For example, the filters and rotations in the encoder are simply the inverse of the operation performed by the decoder. Similarly, the quantizers generally optimize for the mean square error (because noise shaping is part of the bit-stream itself), so no special search is required. For this reason, only the less straightforward aspects of the encoder are described here. </t> <section anchor="pitch-prefilter" title="Pitch prefilter"> <t>The pitch prefilter is applied after the pre-emphasis and before the de-emphasis. It's applied in such a way as to be the inverse of the decoder's post-filter. The main non-obvious aspect of the prefilter is the selection of the pitch period. The pitch search should be optimised for the following criteria: <list style="symbols"> <t>continuity: it is important that the pitch period does not change abruptly between frames; and</t> <t>avoidance of pitch multiples: when the period used is a multiple of the real period (lower frequency fundamental), the post-filter loses most of its ability to reduce noise</t> </list> </t> </section> <section anchor="normalization" title="Bands and Normalization"> <t> The MDCT output is divided into bands that are designed to match the ear's critical bands for the smallest (2.5 ms) frame size. The larger frame sizes use integer multiples of the 2.5 ms layout. For each band, the encoder computes the energy that will later be encoded. Each band is then normalized by the square root of the <spanx style="strong">unquantized</spanx> energy, such that each band now forms a unit vector X. The energy and the normalization are computed by compute_band_energies() and normalise_bands() (bands.c), respectively. </t> </section> <section anchor="energy-quantization" title="Energy Envelope Quantization"> <t> Energy quantization (both coarse and fine) can be easily understood from the decoding process. The quantizer simply minimizes the log energy error for each band, with the exception that at very low rate, larger errors are allowed in the coarse energy to minimize the bit-rate. When the avaialble CPU requirements allow it, it is best to try encoding the coarse energy both with and without inter-frame prediction such that the best prediction mode can be selected. The optimal mode depends on the coding rate, the available bit-rate, and the current rate of packet loss. </t> </section> <!-- Energy quant --> <section anchor="pvq" title="Spherical Vector Quantization"> <t>CELT uses a Pyramid Vector Quantization (PVQ) <xref target="PVQ"></xref> codebook for quantizing the details of the spectrum in each band that have not been predicted by the pitch predictor. The PVQ codebook consists of all sums of K signed pulses in a vector of N samples, where two pulses at the same position are required to have the same sign. Thus the codebook includes all integer codevectors y of N dimensions that satisfy sum(abs(y(j))) = K. </t> <t> In bands where there are sufficient bits allocated the PVQ is used to encode the unit vector that results from the normalization in <xref target="normalization"></xref> directly. Given a PVQ codevector y, the unit vector X is obtained as X = y/||y||, where ||.|| denotes the L2 norm. </t> <section anchor="pvq-search" title="PVQ Search"> <t> The search for the best codevector y is performed by alg_quant() (vq.c). There are several possible approaches to the search, with a trade-off between quality and complexity. The method used in the reference implementation computes an initial codeword y1 by projecting the normalized spectrum X onto the codebook pyramid of K-1 pulses: </t> <t> y0 = round_towards_zero( (K-1) * X / sum(abs(X))) </t> <t> Depending on N, K and the input data, the initial codeword y0 may contain from 0 to K-1 non-zero values. All the remaining pulses, with the exception of the last one, are found iteratively with a greedy search that minimizes the normalized correlation between y and X: <figure align="center"> <artwork align="center"><![CDATA[ T J = -X * y / ||y|| ]]></artwork> </figure> </t> <t> The search described above is considered to be a good trade-off between quality and computational cost. However, there are other possible ways to search the PVQ codebook and the implementers MAY use any other search methods. </t> </section> </section> </section> </section> <section title="Conformance"> <t> It is the intention to allow the greatest possible choice of freedom in implementing the specification. For this reason, outside of a few exceptions noted in this section, conformance is defined through the reference implementation of the decoder provided in <xref target="ref-implementation"/>. Although this document includes an English description of the codec, should the description contradict the source code of the reference implementation, the latter shall take precedence. </t> <t> Compliance with this specification means that a decoder's output MUST be within the thresholds specified by the opus_compare.c tool (included with the code) when compared to the reference implementation. Either the floating-point implementation or the fixed-point implementation can be used as a reference and being within the threshold for one of the two is sufficient. </t> <t> To complement the Opus specification, the "Opus Custom" codec is defined to handle special sampling rates and frame rates that are not supported by the main Opus specification. Use of Opus Custom is discouraged for all but very special applications for which a frame size different from 2.5, 5, 10, or 20 ms is needed (for either complexity or latency reasons). Such applications will not be compatible with the "main" Opus codec. In Opus Custom operation, only the CELT layer is available, which is available using the celt_* function calls in celt.h. </t> </section> <section anchor="security" title="Security Considerations"> <t> Implementations of the Opus codec need to take appropriate security considerations into account, as outlined in <xref target="DOS"/> and <xref target="SECGUIDE"/>. It is extremely important for the decoder to be robust against malicious payloads. Malicious payloads must not cause the decoder to overrun its allocated memory or to take an excessive amount of resources to decode. Although problems in encoders are typically rarer, the same applies to the encoder. Malicious audio streams must not cause the encoder to misbehave because this would allow an attacker to attack transcoding gateways. </t> <t> The reference implementation contains no known buffer overflow or cases where a specially crafted packet or audio segment could cause a significant increase in CPU load. However, on certain CPU architectures where denormalized floating-point operations are much slower than normal floating-point operations, it is possible for some audio content (e.g., silence or near-silence) to cause a certain an increase in CPU load. Denormals can be introduced by reordering operations in the compiler and depend on the target architecture, so it is difficult to guarantee that an implementation avoids them. For architectures on which denormals are problematic, adding very small floating-point offsets to the affected signals to prevent significant numbers of denormalized operations is RECOMMENDED. Alternatively, it is often possible to configure the hardware to treat denormals as zero (DAZ). No such issue exists for the fixed-point reference implementation. </t> <t>The reference implementation was validated in the following conditions: <list style="numbers"> <t> Sending the decoder valid packets generated by the reference encoder and verifying that the decoder's final range coder state matches that of the encoder. </t> <t> Sending the decoder packets generated by the reference encoder and then subjected to random corruption. </t> <t>Sending the decoder random packets.</t> <t> Sending the decoder packets generated by a version of the reference encoder modified to make random coding decisions (internal fuzzing), including mode switching, and verifying that the range coder final states match. </t> </list> In all of the conditions above, both the encoder and the decoder were run inside the <eref target="http://valgrind.org/">Valgrind</eref> memory debugger, which tracks reads and writes to invalid memory regions as well as the use of uninitialized memory. There were no errors reported on any of the tested conditions. </t> </section> <section title="IANA Considerations"> <t> This document has no actions for IANA. </t> </section> <section anchor="Acknowledgments" title="Acknowledgments"> <t> Thanks to all other developers, including Raymond Chen, Soeren Skak Jensen, Gregory Maxwell, Christopher Montgomery, and Karsten Vandborg Soerensen. We would also like to thank Igor Dyakonov, Jan Skoglund, and Christian Hoene for their help with subjective testing of the Opus codec. Thanks to Ralf Giles, John Ridges, Ben Schwartz, Keith Yan, and many others on the Opus and CELT mailing lists for their bug reports and feedback, as well as Ralph Giles, Christian Hoene, and Kat Walsh, for their feedback on the draft. </t> </section> </middle> <back> <references title="Informative References"> <reference anchor='requirements'> <front> <title>Requirements for an Internet Audio Codec</title> <author initials='J.-M.' surname='Valin' fullname='J.-M. Valin'> <organization /></author> <author initials='K.' surname='Vos' fullname='K. Vos'> <organization /></author> <author> <organization>IETF</organization></author> <date year='2011' month='August' /> <abstract> <t>This document provides specific requirements for an Internet audio codec. These requirements address quality, sampling rate, bit-rate, and packet-loss robustness, as well as other desirable properties. </t></abstract></front> <seriesInfo name='RFC' value='6366' /> <format type='TXT' target='http://tools.ietf.org/rfc/rfc6366.txt' /> </reference> <reference anchor='SILK'> <front> <title>SILK Speech Codec</title> <author initials='K.' surname='Vos' fullname='K. Vos'> <organization /></author> <author initials='S.' surname='Jensen' fullname='S. Jensen'> <organization /></author> <author initials='K.' surname='Soerensen' fullname='K. Soerensen'> <organization /></author> <date year='2010' month='March' /> <abstract> <t></t> </abstract></front> <seriesInfo name='Internet-Draft' value='draft-vos-silk-01' /> <format type='TXT' target='http://tools.ietf.org/html/draft-vos-silk-01' /> </reference> <reference anchor="laroia-icassp"> <front> <title abbrev="Robust and Efficient Quantization of Speech LSP"> Robust and Efficient Quantization of Speech LSP Parameters Using Structured Vector Quantization </title> <author initials="R.L." surname="Laroia" fullname="R."> <organization/> </author> <author initials="N.P." surname="Phamdo" fullname="N."> <organization/> </author> <author initials="N.F." surname="Farvardin" fullname="N."> <organization/> </author> </front> <seriesInfo name="ICASSP-1991, Proc. IEEE Int. Conf. Acoust., Speech, Signal Processing, pp. 641-644, October" value="1991"/> </reference> <reference anchor="leblanc-tsap"> <front> <title>Efficient Search and Design Procedures for Robust Multi-Stage VQ of LPC Parameters for 4 kb/s Speech Coding</title> <author initials="W.P." surname="LeBlanc" fullname=""> <organization/> </author> <author initials="B." surname="Bhattacharya" fullname=""> <organization/> </author> <author initials="S.A." surname="Mahmoud" fullname=""> <organization/> </author> <author initials="V." surname="Cuperman" fullname=""> <organization/> </author> </front> <seriesInfo name="IEEE Transactions on Speech and Audio Processing, Vol. 1, No. 4, October" value="1993" /> </reference> <reference anchor='CELT'> <front> <title>Constrained-Energy Lapped Transform (CELT) Codec</title> <author initials='J-M.' surname='Valin' fullname='J-M. Valin'> <organization /></author> <author initials='T.B.' surname='Terriberry' fullname='Timothy B. Terriberry'> <organization /></author> <author initials='G.' surname='Maxwell' fullname='G. Maxwell'> <organization /></author> <author initials='C.' surname='Montgomery' fullname='C. Montgomery'> <organization /></author> <date year='2010' month='July' /> <abstract> <t></t> </abstract></front> <seriesInfo name='Internet-Draft' value='draft-valin-celt-codec-02' /> <format type='TXT' target='http://tools.ietf.org/html/draft-valin-celt-codec-02' /> </reference> <reference anchor='DOS'> <front> <title>Internet Denial-of-Service Considerations</title> <author initials='M.' surname='Handley' fullname='M. Handley'> <organization /></author> <author initials='E.' surname='Rescorla' fullname='E. Rescorla'> <organization /></author> <author> <organization>IAB</organization></author> <date year='2006' month='December' /> <abstract> <t>This document provides an overview of possible avenues for denial-of-service (DoS) attack on Internet systems. The aim is to encourage protocol designers and network engineers towards designs that are more robust. We discuss partial solutions that reduce the effectiveness of attacks, and how some solutions might inadvertently open up alternative vulnerabilities. This memo provides information for the Internet community.</t></abstract></front> <seriesInfo name='RFC' value='4732' /> <format type='TXT' octets='91844' target='ftp://ftp.isi.edu/in-notes/rfc4732.txt' /> </reference> <reference anchor='SECGUIDE'> <front> <title>Guidelines for Writing RFC Text on Security Considerations</title> <author initials='E.' surname='Rescorla' fullname='E. Rescorla'> <organization /></author> <author initials='B.' surname='Korver' fullname='B. Korver'> <organization /></author> <date year='2003' month='July' /> <abstract> <t>All RFCs are required to have a Security Considerations section. Historically, such sections have been relatively weak. This document provides guidelines to RFC authors on how to write a good Security Considerations section. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements.</t></abstract></front> <seriesInfo name='BCP' value='72' /> <seriesInfo name='RFC' value='3552' /> <format type='TXT' octets='110393' target='ftp://ftp.isi.edu/in-notes/rfc3552.txt' /> </reference> <reference anchor="range-coding"> <front> <title>Range encoding: An algorithm for removing redundancy from a digitised message</title> <author initials="G." surname="Nigel" fullname=""><organization/></author> <author initials="N." surname="Martin" fullname=""><organization/></author> <date year="1979" /> </front> <seriesInfo name="Proc. Institution of Electronic and Radio Engineers International Conference on Video and Data Recording" value="" /> </reference> <reference anchor="coding-thesis"> <front> <title>Source coding algorithms for fast data compression</title> <author initials="R." surname="Pasco" fullname=""><organization/></author> <date month="May" year="1976" /> </front> <seriesInfo name="Ph.D. thesis" value="Dept. of Electrical Engineering, Stanford University" /> </reference> <reference anchor="PVQ"> <front> <title>A Pyramid Vector Quantizer</title> <author initials="T." surname="Fischer" fullname=""><organization/></author> <date month="July" year="1986" /> </front> <seriesInfo name="IEEE Trans. on Information Theory, Vol. 32" value="pp. 568-583" /> </reference> </references> <section anchor="ref-implementation" title="Reference Implementation"> <t>This appendix contains the complete source code for the reference implementation of the Opus codec written in C. By default, this implementation relies on floating-point arithmetic, but it can be compiled to use only fixed-point arithmetic by defining the FIXED_POINT macro. Information on building and using the reference implementation is available in the README file. </t> <t>The implementation can be compiled with either a C89 or a C99 compiler. It is reasonably optimized for most platforms such that only architecture-specific optimizations are likely to be useful. The FFT used is a slightly modified version of the KISS-FFT library, but it is easy to substitute any other FFT library. </t> <t> While the reference implementation does not rely on any <spanx style="emph">undefined behavior</spanx> as defined by C89 or C99, it relies on common <spanx style="emph">implementation-defined behavior</spanx> for two's complement architectures: <list style="symbols"> <t>Right shifts of negative values are consistent with two's complement arithmetic, so that a>>b is equivalent to floor(a/(2^b))</t> <t>For conversion to a signed integer of N bits, the value is reduced modulo 2^N to be within range of the type</t> <t>The result of integer division of a negative values is truncated towards zero</t> <t>The compiler provides a 64-bit integer type (a C99 requirement which is supported by most c89 compilers)</t> </list> </t> <t> In its current form, the reference implementation also requires the following architectural characteristics to obtain acceptable performance: <list style="symbols"> <t>two's complement arithmetic</t> <t>at least a 16 bit by 16 bit integer multiplier (32-bit result)</t> <t>at least a 32-bit adder/accumulator</t> </list> </t> <section title="Extracting the source"> <t> The complete source code can be extracted from this draft, by running the following command line: <list style="symbols"> <t><![CDATA[ cat draft-ietf-codec-opus.txt | grep '^\ \ \ ###' | sed 's/\s\s\s###//' | base64 -d > opus_source.tar.gz ]]></t> <t> tar xzvf opus_source.tar.gz </t> <t>cd opus_source</t> <t>make</t> </list> </t> </section> <section title="Development Versions"> <t> The current development version of the source code is available in a <eref target='git://git.opus-codec.org/opus.git'>Git repository</eref>. Development snapshots are provided at <eref target='http://opus-codec.org/'/>. </t> </section> <section title="Base64-encoded source code"> <t> <?rfc include="opus_source.base64"?> </t> </section> </section> <section anchor="self-delimiting-framing" title="Self-Delimiting Framing"> <t> To use the internal framing described in <xref target="modes"/>, the decoder must know the total length of the Opus packet, in bytes. This section describes a simple variation of that framing which can be used when the total length of the packet is not known. Nothing in the encoding of the packet itself allows a decoder to distinguish between the regular, undelimited framing and the self-delimiting framing described in this appendix. Which one is used and where must be established by context at the transport layer. It is RECOMMENDED that a transport layer choose exactly one framing scheme, rather than allowing an encoder to signal which one it wants to use. </t> <t> For example, although a regular Opus stream does not support more than two channels, a multi-channel Opus stream may be formed from several one- and two-channel streams. To pack an Opus packet from each of these streams together in a single packet at the transport layer, one could use the self-delimiting framing for all but the last stream, and then the regular, undelimited framing for the last one. Reverting to the undelimited framing for the last stream saves overhead (because the total size of the transport-layer packet will still be known), and ensures that a "multi-channel" stream which only has a single Opus stream uses the same framing as a regular Opus stream does. This avoids the need for signaling to distinguish these two cases. </t> <t> The self-delimiting framing is identical to the regular, undelimited framing from <xref target="modes"/>, except that each Opus packet contains one extra length field, encoded using the same one- or two-byte scheme from <xref target="frame-length-coding"/>. This extra length immediately precedes the compressed data of the first Opus frame in the packet, and is interpreted in the various modes as follows: <list style="symbols"> <t> Code 0 packets: It is the length of the single Opus frame (see <xref target="sd_code0_packet"/>). </t> <t> Code 1 packets: It is the length used for both of the Opus frames (see <xref target="sd_code1_packet"/>). </t> <t> Code 2 packets: It is the length of the second Opus frame (see <xref target="sd_code2_packet"/>).</t> <t> CBR Code 3 packets: It is the length used for all of the Opus frames (see <xref target="sd_code3cbr_packet"/>). </t> <t>VBR Code 3 packets: It is the length of the last Opus frame (see <xref target="sd_code3vbr_packet"/>). </t> </list> </t> <figure anchor="sd_code0_packet" title="A Self-Delimited Code 0 Packet" align="center"> <artwork align="center"><![CDATA[ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |0|0|s| config | N1 (1-2 bytes): | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | Compressed frame 1 (N1 bytes)... : : | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> <figure anchor="sd_code1_packet" title="A Self-Delimited Code 1 Packet" align="center"> <artwork align="center"><![CDATA[ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|0|s| config | N1 (1-2 bytes): | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ : | Compressed frame 1 (N1 bytes)... | : +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ : | Compressed frame 2 (N1 bytes)... | : +-+-+-+-+-+-+-+-+ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> <figure anchor="sd_code2_packet" title="A Self-Delimited Code 2 Packet" align="center"> <artwork align="center"><![CDATA[ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |0|1|s| config | N1 (1-2 bytes): N2 (1-2 bytes : | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ : | Compressed frame 1 (N1 bytes)... | : +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | Compressed frame 2 (N2 bytes)... : : | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> <figure anchor="sd_code3cbr_packet" title="A Self-Delimited CBR Code 3 Packet" align="center"> <artwork align="center"><![CDATA[ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|1|s| config | M |p|0| Pad len (Opt) : N1 (1-2 bytes): +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : Compressed frame 1 (N1 bytes)... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : Compressed frame 2 (N1 bytes)... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : ... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : Compressed frame M (N1 bytes)... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ : Opus Padding (Optional)... | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> <figure anchor="sd_code3vbr_packet" title="A Self-Delimited VBR Code 3 Packet" align="center"> <artwork align="center"><![CDATA[ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|1|s| config | M |p|1| Padding length (Optional) : +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ : N1 (1-2 bytes): ... : N[M-1] | N[M] : +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : Compressed frame 1 (N1 bytes)... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : Compressed frame 2 (N2 bytes)... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : ... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : Compressed frame M (N[M] bytes)... : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ : Opus Padding (Optional)... | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ]]></artwork> </figure> </section> </back> </rfc>