shithub: opus

590e9ce4 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/11 11:25

Move the common functions from dump_data.c to freq.c

9b581a13 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/11 11:10

Rename denoise.c to dump_data.c, rnnoise.h to freq.h

3cb52ece – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/11 10:37

...

86c15a7a – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/11 08:31

Remove more useless code

4eb2d344 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/11 06:49

Cleanup: Remove useless code (more to come)

37ddc0a8 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/10 21:00

Add -test or -train option

74b98437 – David Rowe <david@rowetel.com> authored and Jean-Marc Valin <jmvalin@jmvalin.ca> committed on 2018/12/10 20:41

Vectorization testing code

03dcb819 – David Rowe <david@rowetel.com> authored and Jean-Marc Valin <jmvalin@jmvalin.ca> committed on 2018/12/10 20:40

Error messages

7dc696b9 – David Rowe <david@rowetel.com> authored and Jean-Marc Valin <jmvalin@jmvalin.ca> committed on 2018/12/10 16:28

refactored for different machines, sgemv_accum16 using NEON intrisics

bc108e9a – David <david@rowetel.com> authored and Jean-Marc Valin <jmvalin@jmvalin.ca> committed on 2018/12/07 06:29

checks that files opened OK

8c271d60 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/10 11:15

Controlling per-gate sparsity

b9e0ea23 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/10 06:23

Fix flooring of the pitch period

d533e402 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/10 06:22

Fix DCT normalization

a02d9c85 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/09 16:21

reduce memory use of training code

2facc08b – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/09 16:20

Avoid rounding the prediction multiple times

3e712482 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/09 11:00

Fix compile

0fb031a9 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/01 10:58

Produce at least ~14 hours of augmented speech

3dcbb012 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/01 10:52

Add variable gain and response

f9337256 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/01 09:38

Chopping silence from the training data

407eec12 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/09 10:45

make dump_data output LPC as converted from features

e0d2b105 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/07 13:16

Compute LPC from features

1dcd5732 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/06 11:27

Do proper saturation

771cc786 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/01 21:10

Support for plain AVX with no FMA

91d90676 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/12/01 07:05

Remove the need for useless exc and pred files

b05f950e – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/30 05:56

Using the right name: s/gemm/sgemv/

c395a68b – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/30 05:46

moving code around

05f4851d – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/30 05:32

Making the code work even without AVX2/FMA

1956467d – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/29 20:55

Add AVX2/FMA to gcc options

d7f0abcd – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/29 15:09

Delaying the softmax() to avoid the pow()

faf3fe3d – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/29 14:50

gemm_accum16() doesn't need a multiple of 16 columns (just lines).

7ee79b63 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/29 14:43

Add AXV versions of exp(), tanh() and sigmoid()

d961d009 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/28 15:20

Managing to actually use sparse matrices

4de3e53a – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/28 13:49

Adding some sparse GRU support

ec671ed9 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/28 09:57

Quick and dirty AVX2 implementation of gemm_accum

15fb1b3c – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/28 09:13

Moving GRU_A's condition computation to the frame rate network

732fce9a – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/28 09:05

Pre-computing GRU_A's input contribution.

040aa437 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/28 07:37

Simper GRU implementation just for reset_after.

6c2f7e58 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/27 10:08

compiling synthesis

36a0bf8c – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/27 09:50

Wow, managed two bugs in a 25-character line

c7b978b9 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/27 09:37

Fix reset_after GRU

3c694db2 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/27 08:11

Better rounding

0ddfdfc7 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/27 07:34

Add deemphasis

5ac0ac7a – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/27 06:43

Add code for copying the LPC

e25a585d – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/26 19:23

Match Python boundary condition

4ccfbdff – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/26 13:41

Frame network seems to be working

538f2556 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/26 11:02

Starting to actually test this -- fix a few OOB reads

8d62ba06 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/26 10:39

...

c0e8f37c – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/26 07:44

Cleaning up the API

575d8d6f – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/26 06:04

Adding sampling

91c9524a – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/25 21:49

Getting there

4cf17741 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/25 21:33

More plumbing

7119eaf3 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/25 12:20

Plumbing for the frame rate network

70fdf474 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/25 08:15

Copyright headers

60c97b97 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/24 11:19

Proper ordering for MDense weights

141830ce – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/24 11:00

Fixing includes

37fbcaee – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/24 10:51

mdense max size

94ac0841 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/24 10:47

Precomputing sizes

c025744e – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/24 10:30

Fix conv1d, default to size 384

66486004 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/24 07:23

Implement MDense

d4046036 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/24 06:32

Dump Conv1D (didn't check weight ordering at all)

477d0873 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/23 18:33

Dump embedding

d93239e9 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/23 15:07

Using non-cudnn version of the GRU for the weights

b0c61158 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/23 14:51

More meaningful names

b9cd61be – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/23 14:43

Work in progress translation to C

8caaa5e9 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/11/06 22:26

Output directly to 16-bit (raw) PCM

cf926d11 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/25 15:19

Project name is LPCNet

bf04b53a – David Rowe <david@rowetel.com> authored and Jean-Marc Valin <jmvalin@jmvalin.ca> committed on 2018/10/25 12:19

Cleanup

db7569c3 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/24 10:57

moving hyper-parameters to new_lpcnet_model() arguments

5ff58fa1 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/24 10:09

remove import that no longer exists

92281bbe – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/23 19:10

s/CELPNet/LPCNet/

7c28191b – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/22 09:40

Rename the current files to use the LPCNet name since they're no longer WaveNet

97dcf52a – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/22 09:40

Remove no longer used files (old wavenet and LPCNet implementations)

3122b6b3 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/20 22:42

most promising model for now

f9fe6c0e – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/16 21:36

clear pitch features in testing too

4e331f37 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/16 16:23

LPC generation from the cepstral coefficients

fb1d4fde – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/16 00:05

...

fa1d2824 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/14 22:45

Add diagonal to sparsity mask

f13debcf – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/14 08:21

Arbitrary 16x1 sparseness

62f330ec – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/15 21:00

better training params

a263f7c1 – David Rowe <david@rowetel.com> authored and Jean-Marc Valin <jmvalin@jmvalin.ca> committed on 2018/10/14 08:36

Adding comments

beaa370a – David Rowe <david@rowetel.com> authored and Jean-Marc Valin <jmvalin@jmvalin.ca> committed on 2018/10/14 08:27

Adding a README.md

d75b51b1 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/13 10:52

Reduce sampling temperature for voiced frames

c74876bb – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/12 23:54

Adding some instructions

f3eb6164 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/12 23:41

Cleaning up the synthesis code

9756feef – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/10 12:48

...

495f8ea5 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/09 21:41

second RNN

36989772 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/10 13:28

add license

03fa20d5 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/09 08:27

remove unused/dead code

a9835c4e – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/08 23:10

more cleanup

aba9af8b – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/08 22:39

mu-law code cleanup

08211c27 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/08 21:01

missing script

ea1391e1 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/03 18:30

deeper features

639766b3 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/02 17:17

pitch embedding

2d74d318 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/10/02 14:26

...

c381db56 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/08/23 20:20

Use excitation as input

8f6e490b – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/08/22 19:32

clean excitation

8a276fb4 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/08/21 15:17

predicting excitation

a922f83c – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/08/21 09:02

Fix input noise

08b5fe6c – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/08/16 20:31

working decoder

87cd75f6 – Jean-Marc Valin <jmvalin@jmvalin.ca> authored on 2018/08/16 18:43

Training seems to work