Commits · fb3112ade604387b27030008ccf9c204efbb19fd · Xiph.Org / Opus

Dec 13, 2018
- Removing most of the full-overlap assumptions from freq.[ch] · fb3112ad
  Jean-Marc Valin authored 6 years ago
  
  fb3112ad
- Missing #include · e8c96212
  Jean-Marc Valin authored 6 years ago
  
  e8c96212
Dec 12, 2018
- Avoiding an infinite loop · 06b9bfa8
  Jean-Marc Valin authored 6 years ago
  
  Thanks to changeforan on Github
  06b9bfa8
Dec 11, 2018
- Moving the frame out of lpcnet.c and into test_lpcnet.c · 7d9affc3
  Jean-Marc Valin authored 6 years ago
  
  7d9affc3
- More cleanup · ca4a8d4f
  Jean-Marc Valin authored 6 years ago
  
  ca4a8d4f
- Get rid of the TRAINING macro · 242198ea
  Jean-Marc Valin authored 6 years ago
  
  242198ea
- Move the common functions from dump_data.c to freq.c · 590e9ce4
  Jean-Marc Valin authored 6 years ago
  
  590e9ce4
- Rename denoise.c to dump_data.c, rnnoise.h to freq.h · 9b581a13
  Jean-Marc Valin authored 6 years ago
  
  9b581a13
- ... · 3cb52ece
  Jean-Marc Valin authored 6 years ago
  
  3cb52ece
- Remove more useless code · 86c15a7a
  Jean-Marc Valin authored 6 years ago
  
  86c15a7a
- Cleanup: Remove useless code (more to come) · 4eb2d344
  Jean-Marc Valin authored 6 years ago
  
  4eb2d344
- Add -test or -train option · 37ddc0a8
  Jean-Marc Valin authored 6 years ago
  
  -train does data augmentation, cuts silence, and outputs the preemphasis -test only outputs the features for resynthesis
  37ddc0a8
- Vectorization testing code · 74b98437
  David Rowe authored 6 years ago and Jean-Marc Valin committed 6 years ago
  
  Signed-off-by: Jean-Marc Valin <jmvalin@jmvalin.ca>
  74b98437
- Error messages · 03dcb819
  David Rowe authored 6 years ago and Jean-Marc Valin committed 6 years ago
  
  Signed-off-by: Jean-Marc Valin <jmvalin@jmvalin.ca>
  03dcb819
- refactored for different machines, sgemv_accum16 using NEON intrisics · 7dc696b9
  David Rowe authored 6 years ago and Jean-Marc Valin committed 6 years ago
  
  Signed-off-by: Jean-Marc Valin <jmvalin@jmvalin.ca>
  7dc696b9
Dec 10, 2018
- checks that files opened OK · bc108e9a
  David authored 6 years ago and Jean-Marc Valin committed 6 years ago
  
  bc108e9a
- Controlling per-gate sparsity · 8c271d60
  Jean-Marc Valin authored 6 years ago
  
  8c271d60
- Fix flooring of the pitch period · b9e0ea23
  Jean-Marc Valin authored 6 years ago
  
  Without the 0.1 bias, the rounding error could cause an offset of -1
  b9e0ea23
- Fix DCT normalization · d533e402
  Jean-Marc Valin authored 6 years ago
  
  d533e402
- reduce memory use of training code · a02d9c85
  Jean-Marc Valin authored 6 years ago
  
  a02d9c85
- Avoid rounding the prediction multiple times · 2facc08b
  Jean-Marc Valin authored 6 years ago
  
  2facc08b
Dec 09, 2018
- Fix compile · 3e712482
  Jean-Marc Valin authored 6 years ago
  
  3e712482
- Produce at least ~14 hours of augmented speech · 0fb031a9
  Jean-Marc Valin authored 6 years ago
  
  0fb031a9
- Add variable gain and response · 3dcbb012
  Jean-Marc Valin authored 6 years ago
  
  3dcbb012
- Chopping silence from the training data · f9337256
  Jean-Marc Valin authored 6 years ago
  
  f9337256
- make dump_data output LPC as converted from features · 407eec12
  Jean-Marc Valin authored 6 years ago
  
  407eec12
Dec 07, 2018
- Compute LPC from features · e0d2b105
  Jean-Marc Valin authored 6 years ago
  
  e0d2b105
Dec 06, 2018
- Do proper saturation · 1dcd5732
  Jean-Marc Valin authored 6 years ago
  
  1dcd5732
Dec 04, 2018
- Support for plain AVX with no FMA · 771cc786
  Jean-Marc Valin authored 6 years ago
  
  771cc786
Dec 01, 2018
- Remove the need for useless exc and pred files · 91d90676
  Jean-Marc Valin authored 6 years ago
  
  91d90676
Nov 30, 2018
- Using the right name: s/gemm/sgemv/ · b05f950e
  Jean-Marc Valin authored 6 years ago
  
  b05f950e
- moving code around · c395a68b
  Jean-Marc Valin authored 6 years ago
  
  c395a68b
- Making the code work even without AVX2/FMA · 05f4851d
  Jean-Marc Valin authored 6 years ago
  
  05f4851d
- Add AVX2/FMA to gcc options · 1956467d
  Jean-Marc Valin authored 6 years ago
  
  1956467d
- Delaying the softmax() to avoid the pow() · d7f0abcd
  Jean-Marc Valin authored 6 years ago
  
  Now at 5x real-time, with all the low-hanging fruit done.
  d7f0abcd
- gemm_accum16() doesn't need a multiple of 16 columns (just lines). · faf3fe3d
  Jean-Marc Valin authored 6 years ago
  
  faf3fe3d
- Add AXV versions of exp(), tanh() and sigmoid() · 7ee79b63
  Jean-Marc Valin authored 6 years ago
  
  Now 3x faster than real-time
  7ee79b63
Nov 29, 2018
- Managing to actually use sparse matrices · d961d009
  Jean-Marc Valin authored 6 years ago
  
  Now 2x real-time!
  d961d009
Nov 28, 2018
- Adding some sparse GRU support · 4de3e53a
  Jean-Marc Valin authored 6 years ago
  
  Still need to properly dump as sparse.
  4de3e53a
- Quick and dirty AVX2 implementation of gemm_accum · ec671ed9
  Jean-Marc Valin authored 6 years ago
  
  Brings us very close to real-time
  ec671ed9