Retrained coarse energy mean and beta coefficients
Making the allocation slightly less insane at very high rate
Using high-band noise less aggressively
Allowing to change the allocation dynamically.
Updates tandem-test for more robust error reporting and additional frame sizes and sample rates.
Fixes a ibudget bust and a divide-by-zero at very low bit-rate
Fixes a fixed-point overflow on 16-bit architectures
Adds a simple dynamic allocation scheme
Preventing bands from being coded at a rate below (for now) 3/8 bit/sample
Include tuning from exp_vorbis_tuning (disabled for now)
Fixes a minor folding bug for N=2 bands in stereo
Add an extra safety bit in intra_decision()
Fixes rare overflow in intra_decision()
Fixes fixed-point overflow in stereo_merge()
Fixes a fixed-point overflow in haar1()
Speed up the bit allocator interpolation
Copy lowband when needed instead of undoing the transforms
Allowing a max of 8 bits to qtheta
More quant_band() reorganisation.
Restructuring in quant_band()
Fixing tests on machines with no __builtin_clz()
Separates stereo_band_mix() into the intensity and MS stereo cases
Added stereo_merge(), which does the renormalisation too
MacOS X typedef fix
Moves the bit-side gain application to the quantizer
Using 6 steps in allocation interpolation
Simplifies vector renormalisation (and using it less)
Fix a minor, but bitstream-affecting bug
Fixes a divide-by-zero in narrowband mode with small frame size
Simplifies convoluted raw bits calls.
Don't allow transients for 2.5 ms frames.
Fixes an index bug in dump_modes
More sharing across different static modes
static modes simplifications for std mode
minor tuning in the mid-high rates
Raw bit init for for architectures with chars larger than 8 bits
Smoothing out some curves
Tuning in the 24 kb/s area
Tuning at very low bit-rate (around 14 kb/s)
Making the allocation matrix a bit smoother
Converting allocation table to 1/32 bit/sample resolution.
Complexity setting now controls two-pass coarse energy quantization
Ignoring lower bands in folding_decision()
Complexity setting now does something again
Minor simplification to interp_bits2pulses()
New implementation of folding_decision()
Fixes some MSVC warnings
Fine energy allocation cleanup
More fine energy tuning, compensation for N=2
New fine energy allocation tuning.
CELT no longer prints to stderr (unless assertions are enabled)
Simplifies the implementation of RESET by placing all the data that needs
Adds _init() functions that do not allocate the states themselves
Comments on encoder and decoder struct contents
Unb0rk fixed-point build
Removing check_decoder()
Decoder state stored in a single allocated object.
Removing check_encoder()
Encoder state now stored in a single allocated object
Simplifies convoluted/idiotic indexing.
More IMDCT de-uglification
Removing out_mem from the encoder state.
Inverse MDCT no longer keeps channels interleaved.
Small suckage reduction in compute_inv_mdcts(). More to come.
Disabling mdct and fft init code with static modes
No longer making copies of static modes
Fixes an off-by-one bug in the static mode's MDCT twiddles
Fixed an old issue where the fixed-point MDCT twiddles were stored as 32-bit values
coarse probability model in static modes too
static modes update for MDCT
Updating dump_modes to include the MDCT and FFT. More work needed.
Updated static modes for new pulse cache.
MSE measurement details (no codec changes)
Making the fine energy allocation code less ugly.
Rework coarse energy coding.
Fixes a stereo bug introduced in the previous commit
Choosing intra frame energy when it's cheaper than inter
Moving intra decision to quant_coarse_energy()
Shuffling code around -- no change to the algorithm
Reorganizing the VBR code
Implemented variable spreading amount in the decoder
Minor simplification to exp_rotation().
eMeans[] quantized to 8 bits
Using the real spectral means instead of the ones
Adaptive fine offset value
More simplifications to N=2 stereo