shithub: opus

Info  •  Files  •  Log  •  Branches

98b8be09 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/11/03 13:28

Vectorize DRED quantization

fa5e960c – Felicia Lim <flim@google.com> authored and Jean-Marc Valin <jmvalin@jmvalin.ca> committed on 2023/11/03 10:53

Match silenced overflow checks in the sse4.1 version

cfeddc49 – Felicia Lim <flim@google.com> authored and Jean-Marc Valin <jmvalin@amazon.com> committed on 2023/10/20 09:24

Silence some overflow checks

1ada7d4d – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/11/02 22:46

Vectorizing sgemv for multiples of 4 with SSE

166a6c8e – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/30 22:14

Fix silly bug in CELT Deep PLC

74c67a8d – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/30 13:14

Fix CELT PLC for single packet between losses

da60266f – Jan Buethe <jbuethe@amazon.de> authored on 2023/11/02 12:52

updated moc method

feb32828 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/30 10:08

Don't try to use models that aren't loaded

62b54643 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/29 20:08

Speed up general case for float matrix multiply

61fb3b16 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2023/10/29 17:19

Don't use reserved identifiers for include guards

d53531d0 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/29 12:33

Update blob loading code

0b755012 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/28 22:38

Use log approximation when possible

4259d354 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/28 22:20

Reusing already-optimized celt_fir()

b22b11a4 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2023/10/28 20:12

Silence some warnings

ddd5669e – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/28 19:33

Pitch and fargan model updates

c99054da – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/26 21:29

Fix CELT deep PLC bugs

ccb244a7 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/24 05:27

cleanup

bc102f5f – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/24 05:19

Slightly more continuous analysis

64236e52 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/20 22:26

Removing more useless code

ef8115bd – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/20 18:07

Stop using tansig_table.h (both copies)

a30c96aa – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/20 17:31

Cleanup

88c58cfa – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/20 13:08

nnet.h no longer needs to #include "vec.h"

1032e47d – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/20 11:12

more cleanup

7f0d456c – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/20 11:05

Remove unneeded functions in nnet.c

4598fe54 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/20 08:54

Quantizing pitchdnn and rdovae weights

290be25b – Jan Buethe <jbuethe@amazon.de> authored on 2023/10/20 10:24

added 16kHz version of opus_compare in python

1accd247 – Jan Buethe <jbuethe@amazon.de> authored on 2023/10/20 10:14

finalized quantization option in export_rdovae_weights.py

88c8b307 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/19 23:28

Doing some unrolling on ARM/Neon

f512c920 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/19 21:33

Unroll the 3x3 convolution case

d720955d – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/19 12:06

Marking RDOVAE layers to quantize

60ac1c6c – Jan Buethe <jbuethe@amazon.de> authored on 2023/10/19 17:54

prepared quantization implementation for DRED

2192e85b – Jan Buethe <jbuethe@amazon.de> authored on 2023/10/19 17:45

restructured osce readme

055c6830 – Jan Buethe <jbuethe@amazon.de> authored on 2023/10/19 17:34

added LACE/NoLACE checkpoint URL

8d43b185 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/18 13:40

Support OPUS_SET_COMPLEXITY() on decoder side

000af034 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/18 09:00

Avoiding work on the PLC update side

d1309dd2 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/17 22:56

Simplifying the DRED/PLC code

8c7c03e5 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/17 21:44

Don't call the libm tanh()

0e397a72 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/17 20:43

Making the build possible without the models

aca04ce2 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/16 21:52

Default to int8 matrix multiplies when available

f82f9d1e – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/16 19:10

oops

e51d3da9 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/16 19:01

Fix tests

828f2553 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/16 18:13

Remove references to kiss99

e7c9bfbb – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/16 18:01

Finish removing LPCNet

ca035ef1 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/16 12:57

Force Deep PLC on when enabling DRED

6471f801 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/14 23:53

update weight blob code

da2121ab – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/14 23:43

Default Deep PLC/DRED to off

6ea9312a – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/14 23:39

Only compile PLC/DRED conditionally

9ed3c7c9 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/14 22:45

Rename ENABLE_NEURAL_FED to ENABLE_DRED

5c24975c – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/14 22:37

Rename NEURAL_PLC to ENABLE_DEEP_PLC

98726c4c – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/14 19:19

Fix PLC in opus_demo when DRED is not present

cbd3a805 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/14 17:19

minor tweaks

0b7c02ca – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/14 13:21

Remove LPCNet from the build

c0f94366 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/14 13:17

Remove FWGAN from the build for now

60f151b8 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/08/04 12:17

Use FARGAN instead of LPCNet in DRED/PLC

35cb8d7f – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/09 22:18

C implementation of FARGAN

9e76a7bf – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/09 20:51

update fargan to match version 45

d1c5b32a – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2023/10/07 14:52

Fix warning from casting between 1D and 2D arrays

58f3647a – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2023/10/07 13:54

Fix misc warnings

81624caf – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2023/10/07 13:45

Silencing alignment warnings on x86 intrinsics

0563d71b – Jan Buethe <jbuethe@amazon.de> authored on 2023/10/07 14:52

updated osce readme

8f9a7e23 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/05 23:11

New model with wider range of bitrates

f0ec990d – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/01 22:23

Switching to neural pitch estimator

da7f4c6c – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/01 21:47

update model

27663d36 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/21 08:20

Using a DenseNet for DRED

8e8edf71 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/10/01 17:34

Remove unneeded (I think) tanh at the end

33adba02 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/30 23:59

First version of pitch DNN C code

966a2d22 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/30 19:43

Code for 2D convolution

f3b86f94 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/29 22:48

Fix model saving

0459a572 – Jan Buethe <jbuethe@amazon.de> authored on 2023/09/29 11:34

updated PitchDNN export script

ce286958 – Jan Buethe <jbuethe@amazon.de> authored on 2023/09/29 11:31

refactoring and cleanup

49014454 – Jan Buethe <jbuethe@amazon.de> authored on 2023/09/29 10:34

fixed type in error message

c5c214df – Jan Buethe <jbuethe@amazon.de> authored on 2023/09/29 10:25

added rudimentary support for dumping nn.Conv2d layers

25c65a0c – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/27 15:46

Fix stats indexing for state

9a7bb764 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/27 09:01

No features skip needed to align pitch features

a6b4fe37 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/27 09:00

Script to compute the groundtruth data using CREPE

217c40d4 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/27 08:57

dump cleanup, change alignment

ae0a140c – Jan Buethe <jbuethe@amazon.de> authored on 2023/09/26 17:42

updated default values in adv_train_fargan.py

a71ba10f – Jean-Marc Valin <jmvalin@jmvalin.ca> authored and Jean-Marc Valin <jmvalin@amazon.com> committed on 2023/08/11 00:06

Fixes stack overflow for some custom modes

41a4c951 – Jan Buethe <jbuethe@amazon.de> authored and Jean-Marc Valin <jmvalin@amazon.com> committed on 2023/09/26 10:35

changed checkpoint format

733a095b – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/25 11:44

Adapting to new data format/model

f38b4a31 – Krishna Subramani <subramani.krishna97@gmail.com> authored and Jean-Marc Valin <jmvalin@amazon.com> committed on 2023/09/24 20:19

Python code for neural pitch

d88dd893 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored and Jean-Marc Valin <jmvalin@amazon.com> committed on 2023/09/24 13:29

Add noise augmentation to pitch dumping

0100cd95 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/23 08:45

Add pitch feature computation

96d89e99 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/15 12:55

Band-limiting the voicing parameter

c4b83ae6 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/25 11:29

RDOVAE model update

574c766c – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/23 13:48

Infinite loops are bad

00580a63 – Jan Buethe <jbuethe@amazon.de> authored on 2023/09/22 07:39

bugfix

aad74fdf – Jan Buethe <jbuethe@amazon.de> authored on 2023/09/22 07:18

pulse-related bugfix

0a92bc5e – Jan Buethe <jbuethe@amazon.de> authored on 2023/09/21 11:01

more lavoce stuff

52c15629 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/15 20:59

Handle the case where the initial state didn't fit

71da9781 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/15 23:18

updated model

b88644b9 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/15 13:27

Quantizing initial state with rdovae too

2ec31cc5 – Jan Buethe <jbuethe@amazon.de> authored on 2023/09/20 15:06

added FARGAN adversarial training script

82f48d36 – Jan Buethe <jbuethe@amazon.de> authored on 2023/09/13 12:57

removed trailing whitespace in fargan

e7beaec3 – Jan Buethe <jbuethe@amazon.de> authored on 2023/09/13 12:31

integrated JM's FFT ada conv

b24c7b43 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/12 18:58

Remove --has-gain and --passthrough-size args

2f8b36d6 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/11 19:28

Add conditioning interpolation, fwconv layer

72c5ea41 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/06 13:15

Only use one frame of pre-loading

108b75c4 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/06 10:30

Randomly double the training sequence length

d54b9fb4 – Jean-Marc Valin <jmvalin@amazon.com> authored on 2023/09/05 08:16

Adds skip connections