FreeDV
/
LPCNet
mirror of https://github.com/drowe67/LPCNet.git
1
0
Fork 0
Code Issues Releases Wiki Activity
368 Commits
17 Branches
4 Tags
59 MiB
Commit Graph

368 Commits (cc0ecbaf7e7e77a541e2ef2fb37c5fcfacac8f49)

Author SHA1 Message Date
David cc0ecbaf7e utils for Librispeech 2020-04-11 16:25:21 +09:30
David ae72068708 option to remove pitch embedding 2020-04-11 13:31:21 +09:30
David fa7475d8ea good results in 200406a, some issues with pitch not changing in females 2020-04-10 14:22:36 +09:30
David cbc7028343 test_lpcnet actually using LPCs 2020-04-06 07:29:08 +09:30
David 94e096a0d6 about to start none/first/lpc first time for real 2020-04-05 12:47:08 +09:30
David a4dd9c2199 added arguments to plot-pulaw 2020-04-05 08:28:26 +09:30
David a2cf8736f5 first pass at script to run none/first/lpc 2020-04-04 20:33:09 +10:30
David b1c4982d5c generalised testing for no/first/LPC predictor 2020-04-04 18:01:24 +10:30
David e46fb89464 experiments with no/first order/LPC/Q in and out of loop predictors 2020-04-01 07:54:46 +10:30
David c8b1ca1bf3 fixed training script, no more clipping 2020-03-29 09:59:16 +10:30
David dac8da4e3b first pass at Codec 2 700C decoder outputing (rough) audio 2020-03-28 10:17:03 +10:30
David 0882dc8481 tool to create packed ulaw samples, and python script to sanity check them with a plot 2020-03-27 16:57:57 +10:30
drowe67 aa6b56808d
Merge pull request #14 from tmiw/master 
Add CMake rule to ensure LPCNet is built with the correct OSX libraries.
 2020-01-12 06:31:57 +11:00
drowe67 392ae405c0
Update README.md 2019-12-01 09:53:40 +10:30
Mooneer Salem 008ca17bd6 Add CMake rule to ensure Codec2 is built with the correct OSX libraries. 2019-11-23 20:20:40 -08:00
drowe67 34049f8317
Merge pull request #13 from drowe67/dr-nnet-read 
lpcnet_dec -r option to load other networks
 2019-11-02 12:22:20 +10:30
David 9bd00c20b9 lpcnet_dec -r option to load other networks 2019-11-02 12:09:20 +10:30
Richard Shaw 1443da9221 Add basic versioning in preparation for release. 2019-10-10 19:26:05 -05:00
drowe67 9bc2c41d7b
Merge pull request #10 from drowe67/dr-vec-compare 
automated tests for LPCNet states & new network for peter and david
 2019-10-09 18:50:33 +10:30
David a00820d7f9 fixed bug that was causing unstable audio - pitch output from quantiser sometimes dropping beneath 40 which embed matrix wasn't trained for 2019-10-09 18:22:32 +10:30
David f8b793f4b0 add peter and david samples 2019-10-07 08:19:53 +10:30
David 64c066c409 Merge branch 'dr-vec-compare' of github.com:drowe67/LPCNet into dr-vec-compare 2019-10-06 07:04:53 +10:30
David 8c3b64b34e fixed bug in codec2_pitch.c that was casuing problems with interpolation/decimation on 190924a. We were training using a discrete PDF of pitch values, so pitch embedding failed when we interpolated pitch during quantisation 2019-10-06 06:59:46 +10:30
David 82ef384370 ramp generator to test decimation/interpolation in quant_feat - no problems found 2019-10-04 14:39:23 +09:30
David 2c56edf2da minor changes while tracking down distortion in 190924a 2019-10-04 10:59:32 +09:30
David d30b7b9cc1 resolved conflicts 2019-10-01 06:23:00 +09:30
David 16d2733c19 ctests working on Ubuntu 16 aand 18, but not 17. It's not ideal, but a way fwd for now 2019-10-01 06:19:08 +09:30
David f811f6127f debugging Travis run of ctests 2019-09-29 11:02:45 +09:30
David 78243190c9 debugging Travis run of ctests 2019-09-29 10:20:43 +09:30
David 57118e3377 debugging Travis run of ctests 2019-09-29 10:10:55 +09:30
David ea160a8b39 debugging Travis run of ctests 2019-09-29 09:56:16 +09:30
David 0c0ba329fc debugging Travis run of ctests 2019-09-29 09:16:42 +09:30
David 3ac5cee235 ctests pasing again 2019-09-29 08:50:57 +09:30
David fb2d791b2e use 190924a tar ball as default 2019-09-29 07:52:02 +09:30
David 69bdcfbd06 cleaned up dump data a little 2019-09-29 07:00:00 +09:30
David 02ecf2ac30 190924a, using original LPCNet voicing est (pitch gain). I have accidentally written over the 190922a network with c2voicing, so will run again 2019-09-24 19:00:44 +09:30
David 49c95df13d cq_16kHz fine, peter a bit spikey, slightly better than 190920c, but not as gd at 190920a. However, quite remarkably - all.wav sounds fine, even outliers like canadian. This is puzzling, as this training material doesn't include resampling, or wianews sample 2019-09-23 06:51:26 +09:30
David 24bc72efef 190920c in between 190215 and 190920a, not quite good enough 2019-09-22 14:45:44 +09:30
David 18c4f2ba33 just peter and david, but cepstrals rather than mags, sounds as good as previous with mags 2019-09-20 11:27:54 +09:30
David a2df6288a9 training script with just Peter and David that produces improvements over the 190215 network 2019-09-20 07:34:34 +09:30
David f86b8da270 about to reconfigure for just vk5apr and david 2019-09-19 05:36:29 +09:30
drowe67 8b78d3f5bc
Merge pull request #12 from ra1nb0w/cmake 
cmake: add DISABLE_CPU_OPTIMIZATION option

Based on Richard's review I'll merge.  Thanks.
 2019-09-02 15:29:41 +09:30
Davide Gerhard 604021c814
cmake: add DISABLE_CPU_OPTIMIZATION option 
add new option to disable cpu flags auto-discovery.
needed by binary package manager because not all architecture has avx.
 2019-08-21 09:29:04 +02:00
drowe67 0d5292c4dd
Merge pull request #11 from ra1nb0w/cmake 
fix cmake configuration file name
 2019-08-19 05:43:11 +09:30
Davide Gerhard b9ac3c74dc
fix cmake configuration file name 
fix installed cmake configuration as requested by freedv-gui
 2019-08-12 18:05:49 +02:00
David 0693c3c3db added option to use codec 2 voicing estimator. This works well on tinytrain.sh with single m and f speaker 2019-08-04 19:03:31 +09:30
David a67030ba51 using 190804a as the mag unittest, to presever network 2019-08-04 15:47:31 +09:30
David 6ac9b1902d finally - high quality from tinytrain on m & f samples, samme quality and loss as 4a3998 2019-08-04 15:18:58 +09:30
David 45d70671d1 added log mag support and test 2019-08-03 08:19:02 +09:30
David 547b6f65f9 added log mag support and test 2019-08-03 08:18:54 +09:30