Commits · 5c3cc55614e2f7eb2f911269d6cde38dcf3633f1 · Xiph.Org / Opus

Nov 10, 2021
- Minor fixes to kiss99 · 5c3cc556
  Timothy B. Terriberry authored 3 years ago and Jean-Marc Valin committed 3 years ago
  
  5c3cc556
- Using KISS99 (taken from Daala) as RNG · 3a475485
  Jean-Marc Valin authored 3 years ago
  
  3a475485
Oct 30, 2021
- Fix the "no lookahead" case · 81229a74
  Jean-Marc Valin authored 3 years ago
  
  81229a74
- Fix potential overflow in training data for large signals · 71a26567
  Jean-Marc Valin authored 3 years ago
  
  71a26567
Oct 25, 2021
- model update · d490fbf5
  Jean-Marc Valin authored 3 years ago
  
  d490fbf5
- more updates to the instructions · ad374be0
  Jean-Marc Valin authored 3 years ago
  
  ad374be0
Oct 21, 2021
- update instructions · 153e43aa
  Jean-Marc Valin authored 3 years ago
  
  153e43aa
- Minor cleanup · 7fdca7f0
  Jean-Marc Valin authored 3 years ago
  
  7fdca7f0
- controllable look-ahead · 0e523aa3
  Jean-Marc Valin authored 3 years ago
  
  0e523aa3
- Don't hardcode the number of bands · a9bf6cee
  Jean-Marc Valin authored 3 years ago
  
  a9bf6cee
- Add noise before sample embedding · b5b1d501
  Jean-Marc Valin authored 3 years ago
  
  Simulates rounding
  b5b1d501
- Bring test_lpcnet.py up-to-date · 3f7756c5
  Jean-Marc Valin authored 3 years ago
  
  3f7756c5
- Dumping 16-bit linear training data · 144b7311
  Jean-Marc Valin authored 3 years ago
  
  144b7311
- auto-detect end-to-end models · a3ef5968
  Jean-Marc Valin authored 3 years ago
  
  a3ef5968
- Add tensorboard logging · d5b6087f
  Jean-Marc Valin authored 3 years ago
  
  d5b6087f
- Adding option to change frame rate network size · b24e53fd
  Jean-Marc Valin authored 3 years ago
  
  b24e53fd
- Combine LAR+L1 regularization · fe7b54c0
  Jean-Marc Valin authored 3 years ago
  
  fe7b54c0
Oct 19, 2021
- Freeze LPCs when quantizing e2e models · 054d984b
  Jean-Marc Valin authored 3 years ago
  
  054d984b
Oct 09, 2021
- Fix non-128 batch sizes · 8cdc8081
  Jean-Marc Valin authored 3 years ago
  
  Avoid hardcoding the batch size in the model
  8cdc8081
Oct 05, 2021
- Making sure we recompile when updating the model · 37c9bd8d
  Jean-Marc Valin authored 3 years ago
  
  37c9bd8d
Oct 04, 2021
- Add lpcnet_compute_single_frame_features() to header · 2a131ff6
  Jean-Marc Valin authored 3 years ago
  
  2a131ff6
- split model downloading away from autogen.sh · be4f70bc
  Jean-Marc Valin authored 3 years ago
  
  be4f70bc
- update instructions · 82c31b4c
  Jean-Marc Valin authored 3 years ago
  
  82c31b4c
- Merge branch 'exp_quant_grub2' · 2275853a
  Jean-Marc Valin authored 3 years ago
  
  2275853a
- New model with quantized GRU B · 444b4370
  Jean-Marc Valin authored 3 years ago
  
  444b4370
- Re-enabling quantized feature dump · 4a7ce81f
  Jean-Marc Valin authored 3 years ago
  
  4a7ce81f
- Hard quantization for training · c5a17a07
  Jean-Marc Valin authored 3 years ago
  
  Also, using stateful GRU to randomize initialization
  c5a17a07
Sep 15, 2021
- single-frame features for -features option · 3b8d64d7
  Jean-Marc Valin authored 3 years ago
  
  3b8d64d7
Sep 02, 2021
- Fix signed-unsigned biases · e4b4613d
  Jean-Marc Valin authored 3 years ago
  
  e4b4613d
- Using 8-bit recurrent weights for GRU B · 51ef273e
  Jean-Marc Valin authored 3 years ago
  
  51ef273e
Sep 01, 2021
- Same as 47, without the sign augmentation · 8783ef00
  Jean-Marc Valin authored 3 years ago
  
  8783ef00
Aug 17, 2021
- Refactoring frame rate network · 9776e8e8
  Jean-Marc Valin authored 3 years ago
  
  9776e8e8
- noise tuning, flip sign of speech signal · 2c9b8474
  Jean-Marc Valin authored 3 years ago
  
  2c9b8474
Aug 14, 2021
- frame-wise features · 80751bad
  Jean-Marc Valin authored 3 years ago
  
  80751bad
Aug 13, 2021
- oops, fix single-frame pitch · 4c6a5e0e
  Jean-Marc Valin authored 3 years ago
  
  4c6a5e0e
- WIP: single-frame inference · 6b4e3c56
  Jean-Marc Valin authored 3 years ago
  
  6b4e3c56
- Remove some useless buffers · 66c29fb6
  Jean-Marc Valin authored 3 years ago
  
  66c29fb6
Aug 04, 2021
- dump_lpcnet.py should work the same for end2end · adc50cab
  Jean-Marc Valin authored 3 years ago
  
  adc50cab
- Don't remove silence from training data · 7331e17e
  Jean-Marc Valin authored 3 years ago
  
  7331e17e
- Add more training options · c5364153
  Jean-Marc Valin authored 3 years ago
  
  c5364153