mirrors/DeepSpeech
2
0
Fork 0
mirror of https://github.com/mozilla/DeepSpeech.git synced 2025-10-26 11:19:39 +00:00
Code Issues Packages Projects Releases Wiki Activity
239 Commits 63 Branches 109 Tags 17 GiB
8e99eeffdd
Commit Graph

43 Commits

Author SHA1 Message Date
Tilman Kamp
8e99eeffdd Fix #284; BiRNN with stride 2 2016-12-19 16:21:39 +01:00
Alexandre Lissy
6ec9cc91d5 Restore checkpoint automatically based on some rules 
Fixes #250
 2016-11-29 14:15:05 +01:00
Kelly Davis
06bb017c15 Fixed #239 2016-11-28 09:44:41 +01:00
Tilman Kamp
e83b49ee3e Fix #176; Training, validation an test each in separate sessions and own graphs 2016-11-24 18:07:50 +01:00
Kelly Davis
9d8a261a1d Revert "Merge pull request #192 from Cwiiis/validation-graph-duplication-rebased" 
This reverts commit 08064e7120, reversing
changes made to 757fc74e1e.
 2016-11-18 15:17:26 +01:00
Alexandre Lissy
58f0609afa Tracking GPU Usage 
Fixes #178
 2016-11-17 11:49:20 +01:00
Chris Lord
77100ed1df Don't duplicate graph to do validation 2016-11-16 15:37:42 +00:00
Kelly Davis
04af407011 Merge pull request #171 from mozilla/issue164 
Fixed #164
 2016-11-11 06:26:27 +01:00
Kelly Davis
8e90c84337 Modified to have a uniform importer API 2016-11-10 17:54:36 +01:00
Kelly Davis
763ce2b329 Fixed #164 2016-11-10 16:14:39 +01:00
Alexandre Lissy
b3f4f40869 Do not remove existing checkpoint dirs, update training paramteters 
This breaks restore feature. We should let code overwrite existing
checkpoints. This fixes #166.

Change training parameters, to address issue #169.
 2016-11-10 15:54:09 +01:00
Kelly Davis
0707ad89d2 Merge branch 'master' into issue109_inputops 2016-11-09 12:26:18 +01:00
Tilman Kamp
fc94a2e532 Merge pull request #146 from mozilla/issue144 
Fix #144; Ability to limit amount of samples in TEDLIUM
 2016-11-09 08:30:21 +01:00
Reuben Morais
6fb628521d Normalize non-ASCII chars in the importers 2016-11-08 17:07:46 -02:00
Tilman Kamp
b1fc959f7a Fix #144; Ability to limit amount of samples in TEDLIUM 2016-11-08 16:27:36 +01:00
Reuben Morais
d989e8de09 Make sure the initializer passed to tf.scan doesn't break the API contract 
We need to make sure the initializer shape matches the return value
of the callable passed to tf.scan.

This also adds an assertion on the shape of labels and the values
in label_lengths that enforces a condition that is needed for
ctc_label_dense_to_sparse to work.
 2016-11-08 12:19:42 -02:00
Alexandre Lissy
fa0fc14439 Verify date of merges when pulling list of changes 
Fixes #141
 2016-11-08 14:11:34 +01:00
Reuben Morais
dba8f219f7 Only start the importers' queue threads after we initialize variables 2016-11-08 09:39:27 -02:00
Chris Lord
6178c31a20 Write a Tensorflow Serving client 2016-11-08 11:45:28 +01:00
Reuben Morais
c0bb34cfd3 Undo extraction of shared DataSets code from importers 2016-11-08 03:51:26 -02:00
Reuben Morais
182e20187a Switch importers to new input pipeline 2016-11-08 02:35:50 -02:00
Reuben Morais
ca98c5aab8 Expose text_to_char_array in util/text.py 2016-11-07 16:12:58 -02:00
Alexandre Lissy
34848dcda7 Switch dependency from xdg to pyxdg 
Fixes #129
 2016-11-07 15:27:01 +01:00
Alexandre Lissy
23635420fe Expose FULL_TRACE logging capability 2016-11-07 11:00:55 +01:00
Chris Lord
45d8d6b271 Derive n_steps from batch_x.shape 2016-11-04 11:34:22 +00:00
Chris Lord
1ad35baf4a Base ds_dataset_path on ds_importer by default 2016-11-04 11:34:22 +00:00
Tilman Kamp
0849efd6da Fix #111; documented and revisited WER calculation 2016-11-01 16:11:04 +01:00
Kelly Davis
33e9912259 Updates suggested in first review of #102 2016-10-27 15:46:52 +02:00
Kelly Davis
6ff340f82a Fixes #101 2016-10-27 05:58:08 +02:00
Alexandre Lissy
117a6234a8 WER automation code 
Fixes #73
 2016-10-24 16:35:45 +02:00
Alexandre Lissy
781be5d9bc Handling of website publication 
Fixes #73
 2016-10-24 16:35:45 +02:00
Alexandre Lissy
a6d4373cff LDC93S1 Simple sample importer 
Fixes #72
 2016-10-20 17:28:55 +02:00
Alexandre Lissy
3046d9f6bb Ensure merge_logs() sorts directory 
Fixes #71
 2016-10-19 18:51:04 +02:00
Kelly Davis
ef19385701 Merge pull request #70 from Cwiiis/dont-delete-ted-data 
Fix #69 - Don't delete TED data archive after extraction
 2016-10-18 10:55:41 -04:00
Chris Lord
8ef23ee618 Fix #69 - Don't delete TED data archive after extraction 2016-10-18 15:54:06 +01:00
Kelly Davis
43303a2199 Fix #67 
WER is calculated using Levenshtein distance on chars, not words
 2016-10-17 12:48:20 -04:00
Kelly Davis
f8c0b57578 Fixed typo in wers 2016-10-17 08:47:27 -04:00
Kelly Davis
a3abc9d92a Merge of pull requests #49, #50, and #52. Fixes issues #2, #4, #11, #12, #46, #47, and #48 2016-10-13 15:15:39 -04:00
Tilman Kamp
f3439b72d5 Fixed dropout handling and other fixes 2016-10-11 14:39:36 +02:00
Tilman Kamp
147ca6bde1 Fix #12; integration of WER in training, validation and test 2016-10-07 17:23:20 +02:00
Tilman Kamp
122c27c6a2 Fix #11; WER support 2016-10-06 17:57:49 +02:00
Kelly Davis
26035f3626 Fix of issue #23 2016-09-26 10:26:15 +02:00
Kelly Davis
9eebe98aa9 Adding CTC to notebook 2016-09-19 06:11:42 +02:00