shithub: opus

ef12c29f – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/09/14 13:04

Update encoder/decoder

405aa7cf – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/09/11 00:13

WIP: training with different alignment

981d06ee – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/09/09 22:18

Refactoring towards multiple offset decoding

a4f7c157 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/09/09 13:03

Stop decimating in the encoder

fdd51eb7 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/09/08 23:51

RepeatVector no longer likes tensor input

8569121f – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/09/07 23:12

RDO-VAE work in progress

b6ac1c78 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/09/23 23:20

FEC hooks in the PLC code

4befd8bb – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/09/23 23:25

Decreasing look-ahead of default model to 1 frame

d45ab6fc – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/09/23 23:19

Move back to tanh for frame rate network

f5c251c5 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/09/19 15:03

Properly align LPC with lookahead in data loader

dd114baf – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/09/15 20:55

Fix causal PLC for models with non-zero lookahead

f3c738d4 – janpbuethe <jbuethe@amazon.de> authored on 2022/09/07 05:03

removed debug prints in dump_lpcnet.py

60450472 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/09/06 20:38

Merge branch 'plc_challenge' into master

340ab308 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/09/06 19:31

model update

920300c5 – janpbuethe <jbuethe@amazon.de> authored and Jean-Marc Valin <jmvalin@amazon.com> committed on 2022/08/31 09:51

Add lpc weighting and model parameter handling

c1da818f – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/05/12 20:43

PLC instructions

ee08ef0a – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/03/17 14:21

Auto-download PLC model weights

27348d1b – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/03/04 11:23

Running new features twice on resync

d2d847bc – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/03/04 11:20

Boosting pitch correlation at inference time

099f0872 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/03/04 11:19

Compensating for DC filter offset during blending

8ca2ccf1 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/25 08:57

fix another warning

a814a623 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/25 08:56

Silencing a warning

e034b109 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/25 08:55

Biasing for overestimating the pitch correlation

05f02aaa – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/24 08:03

Adaptive bias

3982144f – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/23 22:22

Revert PLC state to use Burg in the causal overlap

ea405cc6 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/21 17:52

Add delay-compensation for non-causal PLC

a7b24203 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/21 16:36

Setting PLC options at run-time

0eb9f0bd – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/20 21:34

Updating the DC tracker even during packet loss

3883f3d3 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/19 13:03

Synchronizing DC removal for noncausal PLC

a11e68dd – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/19 12:28

Evening out CPU load for non-causal PLC

022ddc31 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/16 18:09

Enable pitch xcorr refining

86bf0cb9 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/15 22:44

Remove pitch periodicity attenuation

cb4dbe05 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/15 22:28

oops

78e68295 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/15 21:08

Handle DC offset

527841d0 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/15 20:58

Avoiding discontinuities in the RNN update

1ae958ac – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/15 20:01

Make LPCNet frame rate network use swish

d1a14ac5 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/15 19:58

Making state update more robust to discontinuities

2a8bcf4c – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/12 21:51

3-part pitch loss function

2e06c078 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/10 22:20

more emphasis on pitch loss

8e3ce6e8 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/09 09:14

Add no-blending version

b2847687 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/08 22:06

Getting rid of prediction delay

9297fe52 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/08 19:51

Add -plc_file option

7a7913f3 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/07 10:14

cleanup

fd45fba9 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/06 23:01

Reduce look-ahead to 5 ms

5d32ab8f – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/05 20:14

optional bi-directional concealment

dc539a9c – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/05 12:05

WIP non-causal PLC

2e18f0d1 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/04 17:04

Using Burg cepstrum for feature prediction

b93dbfc0 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/12/01 11:09

Adding Burg spectral estimation code

f3bc6bac – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/02 19:26

Avoiding tmp buffer overflows

3e2198c6 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/02 10:00

Learning to predict time t+1 from time t

d816477c – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/01 22:54

Add decay

cd2e568b – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/02/01 19:22

Using lost packet file instead of uniform random

1db1946f – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/31 21:57

Support for biased loss

186fa616 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/31 21:57

oops, fix initialization

4866e632 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/31 20:31

minus stupid bug

32a63fd3 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/31 18:21

WIP: PLC prediction

c45963d4 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/31 13:49

Code for testing PLC models

e1181bca – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/30 12:29

oops, fix band loss

c8cbfa7e – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/28 21:54

Adding feature prediction (totally untested)

42cbb9ed – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/27 21:44

zero delay

5ad75a54 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/27 21:44

Opus compat

6b070506 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/26 21:45

Implement lpcnet_plc_get_size()

227537c9 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/24 19:08

Avoiding more symbol clashes with Opus

2f5b51c9 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/24 18:21

Avoiding symbol clashes with Opus

805fed73 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/24 11:29

Fix warnings

57f56819 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/24 11:22

Add swish activation support

93d6c397 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/24 10:35

Fix prototypes for funcs that take no args

60a009b4 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/19 12:41

Making codebase C90-compliant

0f2b8d4a – Jean-Marc Valin <jmvalin@amazon.com> authored on 2022/01/19 13:06

Remove useless code carried over from Opus

969bd766 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/08/13 23:58

Adding API for PLC

b2b2e226 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/11/15 10:49

Optional code for interpolating the pitch correlation

976729dc – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/11/12 22:01

Fix minor bug in dynamic programming pitch search

48a7e250 – Timothy B. Terriberry <territim@amazon.com> authored and Jean-Marc Valin <jmvalin@amazon.com> committed on 2021/11/10 10:24

Abort download_model.sh on errors.

5c3cc556 – Timothy B. Terriberry <territim@amazon.com> authored and Jean-Marc Valin <jmvalin@amazon.com> committed on 2021/11/10 13:01

Minor fixes to kiss99

3a475485 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/11/10 12:58

Using KISS99 (taken from Daala) as RNG

81229a74 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/29 22:37

Fix the "no lookahead" case

71a26567 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/29 22:35

Fix potential overflow in training data for large signals

d490fbf5 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/25 12:03

model update

ad374be0 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/25 12:01

more updates to the instructions

153e43aa – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/21 06:13

update instructions

7fdca7f0 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/20 19:16

Minor cleanup

0e523aa3 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/20 19:05

controllable look-ahead

a9bf6cee – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/20 13:20

Don't hardcode the number of bands

b5b1d501 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/20 13:03

Add noise before sample embedding

3f7756c5 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/20 00:01

Bring test_lpcnet.py up-to-date

144b7311 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/13 19:18

Dumping 16-bit linear training data

a3ef5968 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/13 18:12

auto-detect end-to-end models

d5b6087f – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/12 22:53

Add tensorboard logging

b24e53fd – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/12 22:44

Adding option to change frame rate network size

fe7b54c0 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/12 22:33

Combine LAR+L1 regularization

054d984b – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/16 13:20

Freeze LPCs when quantizing e2e models

8cdc8081 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2021/10/08 23:20

Fix non-128 batch sizes

37c9bd8d – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/04 23:00

Making sure we recompile when updating the model

2a131ff6 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/04 11:59

Add lpcnet_compute_single_frame_features() to header

be4f70bc – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/04 11:51

split model downloading away from autogen.sh

82c31b4c – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/03 23:24

update instructions

2275853a – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/03 23:14

Merge branch 'exp_quant_grub2'

444b4370 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/03 23:12

New model with quantized GRU B

4a7ce81f – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/03 23:04

Re-enabling quantized feature dump

c5a17a07 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2021/10/03 22:53

Hard quantization for training