| .. |
|
4stage_direct_split_vq.c
|
simplified Makefile a bit
|
2019-03-23 09:17:41 +10:30 |
|
4stage_pred_vq.c
|
predictive VQ
|
2019-03-27 05:00:50 +10:30 |
|
CMakeLists.txt
|
Add install targets for lpcnet_{enc,dec}.
|
2019-04-23 13:22:55 -05:00 |
|
_kiss_fft_guts.h
|
…
|
|
|
arch.h
|
…
|
|
|
causalconv.py
|
…
|
|
|
celt_lpc.c
|
…
|
|
|
celt_lpc.h
|
…
|
|
|
codec2_kiss_fft.h
|
codec 2 pitch estimator integrated with dump_data and working OK. Updated process.sh script to plot before and after.
|
2019-02-01 17:17:20 +10:30 |
|
codec2_pitch.c
|
tried a few approaches to fix rough/voicing errors udirnf vowellls on low pitched sample canadian.wav. Implemented Codec 2 voicing est, but didn't help (perhaps as untrained for this estimator). Tried smoothing LPCNet estimator, but no effect, evententually settled on small +ve bias to LPCNet pitch est
|
2019-02-08 13:09:58 +10:30 |
|
codec2_pitch.h
|
tried a few approaches to fix rough/voicing errors udirnf vowellls on low pitched sample canadian.wav. Implemented Codec 2 voicing est, but didn't help (perhaps as untrained for this estimator). Tried smoothing LPCNet estimator, but no effect, evententually settled on small +ve bias to LPCNet pitch est
|
2019-02-08 13:09:58 +10:30 |
|
common.h
|
Using log approximations
|
2019-01-01 14:37:19 -05:00 |
|
concat.sh
|
added concat.sh script
|
2018-12-16 09:31:50 +10:30 |
|
design_huffman.m
|
Octave scripts for LPCNet visualisation and quantiser development
|
2019-01-26 07:24:23 +10:30 |
|
diff32.c
|
building up test framework to track down clicking bug on Windows
|
2019-05-12 07:29:26 +09:30 |
|
dump_data.c
|
windows likes specific file type flags with fopen
|
2019-05-11 06:14:44 +09:30 |
|
dump_lpcnet.py
|
Use a single u-law embedding
|
2019-01-21 16:52:57 -05:00 |
|
flat.m
|
experiment with synthetic speech to see if cepstrals vary from frame-frame with a stationary, low pitch signal. No variation observed - good :-)
|
2019-02-08 13:57:36 +10:30 |
|
freq.c
|
More work on making freq.[ch] more generic
|
2018-12-13 17:40:05 -05:00 |
|
freq.h
|
Controlling training offset
|
2018-12-13 18:25:21 -05:00 |
|
gatedconv.py
|
wip...
|
2018-07-23 17:05:21 -04:00 |
|
idct.c
|
Warn is no vectorization available and change uint to unsigned in in idct.c
|
2019-04-22 15:16:25 -05:00 |
|
kiss_fft.c
|
…
|
|
|
kiss_fft.h
|
…
|
|
|
linux_v_windows.m
|
building up test framework to track down clicking bug on Windows
|
2019-05-12 07:29:26 +09:30 |
|
linux_v_windows.sh
|
building up test framework to track down clicking bug on Windows
|
2019-05-12 07:29:26 +09:30 |
|
load_f32.m
|
Octave scripts for LPCNet visualisation and quantiser development
|
2019-01-26 07:24:23 +10:30 |
|
lpcnet.c
|
building up test framework to track down clicking bug on Windows
|
2019-05-12 07:29:26 +09:30 |
|
lpcnet.h
|
building up test framework to track down clicking bug on Windows
|
2019-05-12 07:29:26 +09:30 |
|
lpcnet.py
|
Use real features at the chunk edges rather than zeros
|
2019-01-24 14:16:30 -05:00 |
|
lpcnet_dec.c
|
added --infile --outfile as stdin/stdout pipes are tricky under Windows/Wine
|
2019-05-08 06:39:49 +09:30 |
|
lpcnet_dump.c
|
moved lpcnet dump code into an API module
|
2019-02-14 06:59:47 +10:30 |
|
lpcnet_dump.h
|
moved lpcnet dump code into an API module
|
2019-02-14 06:59:47 +10:30 |
|
lpcnet_enc.c
|
added --infile --outfile as stdin/stdout pipes are tricky under Windows/Wine
|
2019-05-08 06:39:49 +09:30 |
|
lpcnet_freedv.c
|
lpc_enc now in API function
|
2019-04-03 06:24:54 +10:30 |
|
lpcnet_freedv.h
|
lpc_enc now in API function
|
2019-04-03 06:24:54 +10:30 |
|
lpcnet_freedv_internal.h
|
lpc_enc now in API function
|
2019-04-03 06:24:54 +10:30 |
|
lpcnet_quant.c
|
renamed and made static a few functions to prevent namespace clashes when linking with Codec 2
|
2019-04-07 07:25:43 +09:30 |
|
lpcnet_quant.h
|
renamed and made static a few functions to prevent namespace clashes when linking with Codec 2
|
2019-04-07 07:25:43 +09:30 |
|
mbest.c
|
renamed and made static a few functions to prevent namespace clashes when linking with Codec 2
|
2019-04-07 07:25:43 +09:30 |
|
mbest.h
|
renamed and made static a few functions to prevent namespace clashes when linking with Codec 2
|
2019-04-07 07:25:43 +09:30 |
|
mdense.py
|
…
|
|
|
nnet.c
|
refactored for different machines, sgemv_accum16 using NEON intrisics
|
2018-12-10 21:28:29 -05:00 |
|
nnet.h
|
Adding some sparse GRU support
|
2018-11-28 18:49:19 -05:00 |
|
opus_types.h
|
…
|
|
|
pitch.c
|
…
|
|
|
pitch.h
|
…
|
|
|
plot_features.m
|
plot_features.m used to explore why dynamic range of dctLy[0] (first cepstral) is so high
|
2019-02-03 09:50:49 +10:30 |
|
plot_pitch_test.m
|
codec 2 pitch estimator integrated with dump_data and working OK. Updated process.sh script to plot before and after.
|
2019-02-01 17:17:20 +10:30 |
|
plot_speech_pitch.m
|
fixed usage 2 case
|
2019-02-12 18:20:40 +10:30 |
|
plot_speech_quant.m
|
added PNGs of pitch and quantiser contours
|
2019-01-30 20:56:40 +10:30 |
|
plot_train.py
|
plot loss as well
|
2019-02-09 10:28:30 +10:30 |
|
plot_train.sh
|
plot loss as well
|
2019-02-09 10:28:30 +10:30 |
|
plot_wo.m
|
typo
|
2019-01-30 20:54:58 +10:30 |
|
plot_wo_test_ext.m
|
scripts to compare LPCNet internal pitch est to an external (Codec 2) pitch est. Resolved delay as due to quant_feat buffering, and found a bug in quant_feat
|
2019-01-31 19:20:28 +10:30 |
|
process_features.m
|
Octave scripts for LPCNet visualisation and quantiser development
|
2019-01-26 07:24:23 +10:30 |
|
quant2c.c
|
quant_test working with compiled VQ table
|
2019-02-17 22:00:06 +10:30 |
|
quant_dec.c
|
direct or pred at init time
|
2019-04-02 20:00:51 +10:30 |
|
quant_enc.c
|
refactoring encoder
|
2019-04-03 05:31:52 +10:30 |
|
quant_feat.c
|
direct-split integrated into lpcnet_enc/lpcnet_dec, and it is indeed better than the previous pred scheme with bit errors. Stands up to 1% BER OK, few pops and crackles
|
2019-03-23 19:54:39 +10:30 |
|
quant_test.c
|
simplified Makefile a bit
|
2019-03-23 09:17:41 +10:30 |
|
tansig_table.h
|
Work in progress translation to C
|
2018-11-23 19:43:58 -05:00 |
|
tcodec2_pitch.c
|
tried a few approaches to fix rough/voicing errors udirnf vowellls on low pitched sample canadian.wav. Implemented Codec 2 voicing est, but didn't help (perhaps as untrained for this estimator). Tried smoothing LPCNet estimator, but no effect, evententually settled on small +ve bias to LPCNet pitch est
|
2019-02-08 13:09:58 +10:30 |
|
tdump.c
|
windows likes specific file type flags with fopen
|
2019-05-11 06:14:44 +09:30 |
|
test_lpcnet.c
|
building up test framework to track down clicking bug on Windows
|
2019-05-12 07:29:26 +09:30 |
|
test_lpcnet.py
|
Use real features at the chunk edges rather than zeros
|
2019-01-24 14:16:30 -05:00 |
|
test_vec.c
|
Vectorization testing code
|
2018-12-11 01:41:27 -05:00 |
|
train_lpcnet.py
|
3rd arg set prefix of .h5 file, removed pitch gain (voicing) from model
|
2019-02-09 10:28:01 +10:30 |
|
ulaw.py
|
mu-law code cleanup
|
2018-10-09 02:39:12 -04:00 |
|
vec.h
|
refactored for different machines, sgemv_accum16 using NEON intrisics
|
2018-12-10 21:28:29 -05:00 |
|
vec_avx.h
|
refactored for different machines, sgemv_accum16 using NEON intrisics
|
2018-12-10 21:28:29 -05:00 |
|
vec_neon.h
|
refactored for different machines, sgemv_accum16 using NEON intrisics
|
2018-12-10 21:28:29 -05:00 |
|
weight.c
|
weighting utility for training VQs
|
2019-02-12 18:24:16 +10:30 |
ed9494dead