Oops, thanks to Mark Harris for spotting this!
opus_packet_parse_impl() now computes the packet size with padding
Rejects bad multistream frame length
Fixes PLC for sizes that don't match basic Opus frame sizes.
Make -loss valid for opus_demo -e
Fix 40/60ms zero-length frame decode failure
Cleaning up multistream packet validation
Correct opus_packet_parse code 0 packet tests.
Do up-front validation of multistream packets
Removes useless VARDECL() of "out"
Replaces log2() with celt_log2() and fixes a potential divide-by-zero
Fixes scaling of downmix_float() for fixed-point.
Don't use --dirty in genversion.bat.
Merge branch 'exp_surround1'
Pass --always to git describe.
Avoid using C++ keyword 'new'
Avoids unnecessary copying in opus_encode for delayed decision
Makes analysis work when encoding more than 20 ms at a time
run_analysis() doesn't need to return the frame size anymore
Remove trailing whitespace from the license headers.
Makes the fixed-point build only run the float analysis at complexity 10
Moves frame size selection back out of opus_encode_native()
Very basic surround rate calibration
Allowing surround to use lower bandwidth than fullband
More surround masking tuning
Tuning the surround masking
Surround analysis should now work for non-20ms frame sizes.
Code for handling upsampling in surround_analysis()
Adds surround masking to SILK
Fixes delayed decision for fixed-point
Makes speech/music detection work with FIXED_POINT (code still float)
First pass at making the analysis code run with FIXED_POINT
Moves the analysis back to opus_encode_native() to it runs for all streams.
Adds trim and dynalloc control to surround masking
Makes surround_analysis() work in fixed-point
Surround masking rewrite
Implements fixed-point silk_LPC_analysis_filter() in terms of celt_fir()
Converts the fixed-point silk_burg_modified() to use celt_pitch_xcorr()
Share auto-correlation code between SILK and CELT
Makes the SILK pitch search use celt_pitch_xcorr()
Fix more VS2010 header paths.
Fix relative paths the header file in the VS2010 projects.
Fixes a denorm issue in the VoIP mode high-pass filter
Update the RTP draft for resubmit
Fix broken URL in entdec.c.
Better handling of the dynalloc cap for CBR/CVBR
Fixes an integer wrap-around due to an exceedingly large QCONST16
Add SET_MAX_BANDWIDTH to multistream encoder ctl list
Disable MSVC posix security warnings in the project file.
Conditionalize #pragma warn for MSVC.
Revert "Fixes a number of double promotions and missing casts."
Revert "Additional MSVC warning missed in the last commit."
Additional MSVC warning missed in the last commit.
Fixes a number of double promotions and missing casts.
Always try to update the version when autogen.sh is run
Update draft for the Opus RFC
Updates version.mk for 1.1-beta until we can get rid of it
Fixes a denorm problem when the input goes silent after active audio
Fixes build when DISABLE_FLOAT_API is defined
Better handling of centre for surround masking, bitrate tuning
Fixes silly fixed-point bug with the surround downmix
Prevents zero-energy on LFE
More speech/music detection comments (delayed decision)
Commenting the speech/music Markov code
Disables temporal VBR for LFE
Adds some checks to the surround API
Add a couple minor additional decoder tests.
Braces go on the next line.
Brace a number of if statements instead of one-lining them.
Fixes some return without va_end in the api, adds tests.
Make the parallel test mode of automake 1.13 useable
Fixes more MSVC warnings
Avoid name clash with y0 Bessel function
Updates the Makefile.am EXTRA_DIST for the MSVC reorg
Big squashed commit of Garf's MSVC reorg:
Adds "temporal VBR", which should be mostly useful at low bitrate.
Makes dual_inner_prod() more generic to increase its use
Enable SSE for MSVC
Only enable non-multiple of 4 in comb_filter_const() when CUSTOM_MODES is on
Use __declspec(dllexport) on mingw build.
SSE optimization of comb_filter()
Avoids symbol clashes with Speex (pitch_xcorr) and libm (y1)
Converts denormalise_bands() to use 16-bit multiplications
Moves log2Amp inside denormalise_bands() and get rid of bandE[]
Don't call denormalise_bands() on silence
Split cwrsi() by pulses vs. dimensions.
Further speedup in cwrsi() by using the special case for n=2
Further optimization to cwrsi()