Malar Kannan
|
4dd4bb5963
|
implemented phoneme segmented training on samples
|
2017-12-28 18:53:54 +05:30 |
|
Malar Kannan
|
507da49cfa
|
added voicerss tts support for test data generation
|
2017-12-26 14:32:56 +05:30 |
|
Malar Kannan
|
91fde710f3
|
completed the segmentation model
|
2017-12-07 15:17:59 +05:30 |
|
Malar Kannan
|
c1801b5aa3
|
implented segment tfrecords batch data-generator
|
2017-12-07 11:48:19 +05:30 |
|
Malar Kannan
|
c0369d7a66
|
Merge branch 'master' of ssh://gpuaws/~/repos/speech_scoring
|
2017-12-06 17:33:27 +05:30 |
|
Malar Kannan
|
bcf1041bde
|
created segment sample tfrecord writer
|
2017-12-06 17:32:26 +05:30 |
|
Malar Kannan
|
3f76207f0d
|
using pitch contour instead of spectrogram
|
2017-12-04 19:15:17 +05:30 |
|
Malar Kannan
|
20b2d7a958
|
updated model data
|
2017-11-27 14:08:01 +05:30 |
|
Malar Kannan
|
43d5b75db9
|
removing spec_n counter
|
2017-11-24 11:06:42 +00:00 |
|
Malar Kannan
|
235300691e
|
find spec_n from tfrecords
|
2017-11-24 14:26:36 +05:30 |
|
Malar Kannan
|
6355db4af7
|
adding missing model-dir for training constants copying
|
2017-11-22 15:04:02 +05:30 |
|
Malar Kannan
|
1f60183ab8
|
Merge branch 'master' of ssh://invnuc/~/Public/Repos/speech_scoring
|
2017-11-22 14:45:35 +05:30 |
|
Malar Kannan
|
e7fc607578
|
trying mfcc instead of spectrogram
|
2017-11-22 14:45:08 +05:30 |
|
Malar Kannan
|
d2a075422c
|
copying constantws to models
|
2017-11-20 15:15:27 +05:30 |
|
Malar Kannan
|
a5d4ede35d
|
finding number of record by streaming-onepass
|
2017-11-20 12:07:13 +05:30 |
|
Malar Kannan
|
6ff052be9b
|
fixed randomize pair picking
|
2017-11-17 11:57:38 +05:30 |
|
Malar Kannan
|
7fc89c0853
|
1. fixed pairing and data duplicates
2. clean-up
|
2017-11-16 23:41:38 +05:30 |
|
Malar Kannan
|
a9b244a50c
|
the pair generation order is randomized
|
2017-11-15 14:43:39 +05:30 |
|
Malar Kannan
|
e9f54c7f6f
|
1. tuned batchsize
2. fixed last batch carry-over
|
2017-11-15 14:14:17 +05:30 |
|
Malar Kannan
|
10b024866e
|
implemented evaluation of test data with model by overfitting on smaller dataset
|
2017-11-14 17:54:44 +05:30 |
|
Malar Kannan
|
e4b8b4e0a7
|
visualizing and playing sound files where prediction fails
|
2017-11-13 19:22:30 +05:30 |
|
Malar Kannan
|
988f66c2c2
|
avoiding same voice similar variants
|
2017-11-13 17:33:37 +05:30 |
|
Malar Kannan
|
bb72c4045e
|
trying to overfit the model to identify false-negative types
|
2017-11-10 17:52:21 +05:30 |
|
Malar Kannan
|
1190312def
|
removed tfrecord tensor code and remnants
|
2017-11-10 14:15:12 +05:30 |
|
Malar Kannan
|
e9b18921ee
|
implemented train/test split at word-level and generator returns one-shot validation data
|
2017-11-10 14:07:31 +05:30 |
|
Malar Kannan
|
ab452494b3
|
implemented streaming tfreccords
|
2017-11-09 20:31:29 +05:30 |
|
Malar Kannan
|
0a4d4fadeb
|
implemented random sampling of data for oneshot loading
|
2017-11-09 15:00:17 +05:30 |
|
Malar Kannan
|
b3a6aa2f6a
|
clean-up
|
2017-11-08 11:08:19 +05:30 |
|
Malar Kannan
|
7cbfebbf1a
|
1. fixed missing wrong pairs
2.using different progress bakend
|
2017-11-07 17:27:09 +05:30 |
|
Malar Kannan
|
b8a9f87031
|
implemented padding and pipeline is complete
|
2017-11-07 15:18:04 +05:30 |
|
Malar Kannan
|
41b3f1a9fe
|
dropping invalid csv entries
|
2017-11-07 12:43:17 +05:30 |
|
Malar Kannan
|
55e2de2f04
|
using csv writer instead as comma in phrases are mis-aligning columns
|
2017-11-07 11:56:09 +05:30 |
|
Malar Kannan
|
15f29895d4
|
implemented tfrecord reader and model refactor wip
|
2017-11-07 00:10:23 +05:30 |
|
Malar Kannan
|
046343680e
|
implemented siamese pair tfrecord writer
|
2017-11-06 15:48:38 +05:30 |
|
Malar Kannan
|
c187fbe1ca
|
implemented tfrecord writer for spectrograms
|
2017-11-06 14:12:09 +05:30 |
|
Malar Kannan
|
fabd882664
|
tfrecords wip
|
2017-11-06 12:36:20 +05:30 |
|
Malar Kannan
|
5ff437b095
|
computing spectrogram for existing files
|
2017-11-06 12:15:12 +05:30 |
|
Malar Kannan
|
22d353f101
|
skipping missing files
|
2017-11-03 15:20:31 +05:30 |
|
Malar Kannan
|
b4ceeb4eed
|
generating spectrogram parallelly
|
2017-11-03 14:19:19 +05:30 |
|
Malar Kannan
|
6fbf06814c
|
updated model to use dense classifier
|
2017-10-31 13:31:31 +05:30 |
|
Malar Kannan
|
49e6a46efd
|
code cleanup
|
2017-10-26 12:48:31 +05:30 |
|
Malar Kannan
|
5824158af2
|
1. fixed neg values in spectrogram
2. refactored get word spectrogram code
|
2017-10-25 16:52:45 +05:30 |
|
Malar Kannan
|
f1e82a2539
|
added code to record and generate spectrogram, wip test model
|
2017-10-25 15:38:03 +05:30 |
|
Malar Kannan
|
82d0398d2c
|
formatted
|
2017-10-25 13:36:41 +05:30 |
|
Malar Kannan
|
03d49d83e7
|
updated learning rate
|
2017-10-23 20:21:44 +05:30 |
|
Malar Kannan
|
6f3bca61cf
|
1. fixed dimension issue in data
2. experimenting with different base network
|
2017-10-23 19:00:27 +05:30 |
|
Malar Kannan
|
e865f17a0d
|
pickling intermediate data to save memory usage
|
2017-10-20 12:52:11 +05:30 |
|
Malar Kannan
|
b3755ad80e
|
updated tested pickling
|
2017-10-17 19:17:44 +05:30 |
|
Malar Kannan
|
88edcdd239
|
seprated spectrogram generation code
|
2017-10-17 19:11:04 +05:30 |
|
Malar Kannan
|
51a6d6e804
|
added who data method
|
2017-10-17 19:04:07 +05:30 |