Now that we're preventing huge gains, we no longer need to pretend there's
CBR: lock the gain on a subframe when the number of pulses stops going down
merge back in the LTP gain safety limiter
CBR: Using the same gain increase for fixed-point as for float
Oops, edited the wrong comment in the previous commit
Re-tuning compute_equiv_rate()
Getting rid of the SILK penalty for hybrid CBR.
Improving CBR for hybrid mode
Making the encoder more aggressive at using wider bandwidths
comment about reduced codelength component in LTP quantizer error metric
avoid double if statement for finding best quantized LTP coefs
increase NSQ decision delay from 32 to 40
reduced noise floor in noise shaping analysis by 2 dB
increase order of noise shaping filter
more conservative scaling of LTP corrs; remove assert that checks for negative residual energy; discard CB entries leading to such negative energies
Clean up: alignment of comments
Clean up: replace tabs by spaces
removed prefilter
slight clean up
undo some loop unrolling
pitch analysis (in FIX) now scales the input down at the start, instead of at every stage
minor clean up of pointer offsets
simplified computation of LTP coefs
Attenuate SILK PLC gain only for unvoiced speech
We should now be able to be less conservative when using CBR
Makes the encoder more aggressive about meeting the rate target
Using more fine-grained complexity settings for SILK
Fixing bandwidth selection hysteresis
Boosting the high-band bitrate on transients for hybrid mode
Fixes signed integer overlof in silk_ADD_POS_SAT32()
Turn on DTX only when the noise is sufficiently quiet
New DTX that works in all modes (SILK/CELT/HYBRID)
Skip analysis if the frame contains only digital silence
NLSF_VQ_MAX_SURVIVORS is no longer useful
take advantage of more efficient NLSF quantization by reducing number of survivors in search
NLSF_del_dec_quant minor clean up
NLSF_encode now also uses NLSF weights from table
NLSF VQ now uses absolute error, and prediction
NLSF decoding now uses tables for NLSF weights
Implement sum_sqr_shift() using two passes with no branch inside the loops
Simplifying fast_atan2f()
Quality: Increase CELT rate for voiced frames in hybrid mode
Better modeling of the effect of CBR and complexity
Re-enabling high-band attenuation for hybrid mode
Quality: Retrained classifier MLP with better data
Using "hybrid" flag instead of "start!=0"
Quality: Making the encoder more aggressive at using wider bandwidths
Quality: Removes high-band attenuation in hybrid mode -- at least for now
Quality: Adds SILK rate interpolation function
Quality: Makes real CELT VBR work for hybrid too
Quality: Forces trim to 5 on hybrid mode
Quality: Encoding music in full-band stereo from 32 kb/s
Quality: removes VBR attenuation at low bitrate
Slightly increase the safety margin for opus_pcm_soft_clip()
Fixes opus_pcm_soft_clip() returning values slightly larger than +/-1 with -ffast-math
Apply 02f3ac225f99 fix for divide-by-zero to the SSE4.1 version of silk_burg_modified()
Fix RTCD on ARM with Neon Intrinsics but not ASM.
bump static PACKAGE_VERSION string in version.mk
Packets indicating DTX may be up to 2 bytes
Test common configure switches on travis.
Add a minimal travis continuous integration script.
Makes CBR calculations more accurate for 60-ms frames
Respect CBR even at ridiculourly low rate
Prevents an invalid ToC to be generated at ridiculously low rate
make len==0 return OPUS_INVALID_PACKET rather than OPUS_BAD_ARG for opus_packet_parse_impl()
Check for len<1 in opus_packet_parse_impl()
Remove tabs from source code.
Add Aarch64 intrinsic for SIG2WORD16.
Add Aarch64 intrinsics for saturated add/subtract.
Add configure check for Aarch64-specific Neon intrinsics.
Clean up formatting of configure output for ARM intrinsics detection.
Add OPUS_FAST_INT64 definition of silk_SMULWT.
Explicitly cast results of silk OPUS_FAST_INT64 macros back to opus_int32.
Add OPUS_FAST_INT64 flavors of celt/fixed_generic.h macros.
Move OPUS_FAST_INT64 definition to celt/arch.h.
Clean up some intrinsics-related wording in configure.
Enable intrinsics by default.
Add Neon fixed-point implementation of xcorr_kernel.
Add Neon intrinsics for Silk noise shape feedback loop.
Apply Neon short prediction optimization to silk_noise_shape_quantizer_del_dec.
Add Neon intrinsics for Silk noise shape quantization.
Add named constants for ARM architecture variants.
Enable Neon intrinsics for aarch64.
Rename OPUS_ARM_NEON_INTR AM_CONDITIONAL as HAVE_ARM_NEON_INTR, for consistency with x86.
Reorganize ARM CPU #ifdefs.
Move ARM-specific macro overrides to arm-specific file.
Update links in release checklist.
Add checksums for opusfile 0.8.
Replace another useless 32x16 multiply with 16x16
Replacing some 32x16 multiplies with 16x16 when both operands are 16 bits
Fix folding bug in draft too
Assert non-overlapping memcpy. r=mark4o
Forgot to call RESTORE_STACK in 328953e1896432bc700b9b5c69ae2987c5d752f4
Making calls to opus_packet_pad() on a bad packet return OPUS_INVALID_PACKET
minor update draft update