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
1,930 Commits 63 Branches 109 Tags 17 GiB
cdd4530e66
Commit Graph

1930 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Reuben Morais
cdd4530e66 Update README 2019-09-04 16:52:56 +02:00
Reuben Morais
935ede3f28
Merge pull request #2329 from mozilla/tf_upgrade_v2 
Run tf_upgrade_v2 to ease eventual transition to TF 2.0
 2019-08-28 22:00:23 +02:00
Reuben Morais
85d646350f Update name of audio ops package in TF 1.14/TF 2.0 2019-08-28 18:16:13 +02:00
Reuben Morais
670e06365e Run tf_upgrade_v2 on our code 2019-08-28 17:53:24 +02:00
Reuben Morais
06dee673c7
Merge pull request #2327 from mozilla/abseil-flags 
Switch from deprecated tfv1.app to absl-py
 2019-08-28 17:22:20 +02:00
Reuben Morais
24bcdeb3d6 Switch from deprecated tfv1.app to absl-py 2019-08-28 10:55:33 +02:00
Reuben Morais
289e346a66
Merge pull request #2324 from bjornbytes/c-compat 
Make deepspeech.h compilable as C
 2019-08-28 10:38:15 +02:00
bjorn
f7fc74c078 rm trailing whitespace; 2019-08-27 15:38:39 -07:00
bjorn
73b2bbe8da Use struct typedefs for C compatibility; 2019-08-27 15:34:03 -07:00
bjorn
5d24f19115 Only use extern C when compiling as C++; 2019-08-27 15:25:30 -07:00
Reuben Morais
7e96961e35
Merge pull request #2322 from rcgale/master 
Fixed issue where multiple csvs could not load
 2019-08-26 22:46:50 +02:00
Robert Gale
05448441d3 Fixed issue where multiple csvs could not load 
With the new `create_dataset` approach introduced by PR #2283 (read: mine, sorry!), duplicate
indices in the df would cause a fatal error where the columns could not be referenced by
name. Adding `ignore_index=True` during append allows pandas to assign new indices to
rows, and fixes the issue.
 2019-08-26 13:37:05 -07:00
Reuben Morais
4c14c6b78b
Merge pull request #2303 from mozilla/simplify-decoder 
Simplify decoder impl by making it object oriented, avoid pointers where possible
 2019-08-26 17:02:39 +02:00
Reuben Morais
89f63dcd69 Make Scorer init fallible and check it in callers 2019-08-26 13:59:15 +02:00
Reuben Morais
47b9b71776 Automatically format BUILD file with buildifier 2019-08-26 12:01:26 +02:00
Reuben Morais
c730102867 Avoid rebuilding decoder sources for every binary target 2019-08-26 12:01:26 +02:00
Reuben Morais
1b18494e63 Address review comments 2019-08-23 12:19:32 +02:00
Reuben Morais
b2ef9cca83 Make Alphabet init fallible and check it in model creation 2019-08-23 12:19:32 +02:00
Reuben Morais
4dabd248bc Make Alphabet copyable and default-constructable and avoid pointers 2019-08-23 12:19:32 +02:00
Reuben Morais
4d882a8aec Simplify decoder impl by making it object oriented, avoid pointers where possible 2019-08-23 12:19:32 +02:00
Reuben Morais
f442b69aeb
Merge pull request #2320 from mozilla/swig-4.0.1 
Use globbing instead of hardcoding SWIG version
 2019-08-23 12:19:03 +02:00
Reuben Morais
43b60a621c Use globbing instead of hardcoding SWIG version 2019-08-23 10:45:13 +02:00
lissyx
5fa6d23782
Merge pull request #2316 from lissyx/bump-v0.6.0-alpha.5 
Bump VERSION to 0.6.0-alpha.5
 2019-08-22 09:53:46 +02:00
Alexandre Lissy
97c373a8a9 Bump VERSION to 0.6.0-alpha.5 2019-08-22 09:53:17 +02:00
lissyx
18281b9e89
Merge pull request #2311 from lissyx/armbian-buster 
Move to ARMbian Buster
 2019-08-22 09:31:51 +02:00
Alexandre Lissy
dfe8be30b4 Move to ARMbian Buster 
Fixes #2310
 2019-08-21 22:58:10 +02:00
lissyx
d5544b4a15
Merge pull request #2313 from lissyx/static-libsox 
Statically link libsox
 2019-08-21 22:53:18 +02:00
lissyx
1ec34190a8
Merge pull request #2309 from carlfm01/wpf-build 
Add WPF example build
 2019-08-21 22:24:38 +02:00
Alexandre Lissy
8534c0f93a Statically link libsox 2019-08-21 21:35:08 +02:00
Carlos Fonseca
4812276d88 Add WPF example build 2019-08-21 19:04:20 +00:00
lissyx
50e2a99316
Merge pull request #2308 from lissyx/remove-prealloc 
Remove ununsed prealloc frames
 2019-08-21 14:53:47 +02:00
lissyx
0e47048f9c
Merge pull request #2307 from lissyx/rpi-buster 
Build for Raspbian Buster
 2019-08-20 21:49:51 +02:00
Alexandre Lissy
81b3b159c4 Remove ununsed prealloc frames 
Fixes #2298
 2019-08-20 18:38:46 +02:00
Alexandre Lissy
e06fce51ac Move Raspbian support to Buster 
Fixes #2272
 2019-08-20 16:37:26 +02:00
Reuben Morais
b25de5ac05
Merge pull request #2302 from mozilla/issue2294 
Only update time step of leaf prefixes (Fixes #2294)
 2019-08-20 12:04:35 +02:00
Reuben Morais
e3bf5d3cc6 Only update time step of leaf prefixes 
The intention of this check is to improve the accuracy of the timings by recording the time step where the character saw its highest probability rather than the first time step where it was seen. The problem happens when updating the time step of a prefix that already has children. In that case, if any of the children have a time step that is earlier than `new_timestep`, it'll break the linearity of the timings. My fix is to simply check that the prefix we're updating is a leaf.

For example, say during decoding we have the following beams (format is `(char | time)`, tree node id below, nodes with same id are the same object):

```
1. (-1 | 0 ) -> ('s' | 10) -> ('h' | 13) -> ('e' | 14)
        A                B                  C                D

2. (-1 | 0 ) -> ('s' | 10) -> ('h' | 14)
        A                B                  E
```

And the prefix list is [B, C, D, E]. Currently, if we process character 'h' in time step 15 with a probability higher than both C and E, we update both nodes to have time step 15, which breaks linearity in beam 1. With my fix, we only update node E, which is a leaf. In my tests this does fix the problem, but since we don't have any known good quality data to verify against, it's hard to know if it has other side effects.
 2019-08-20 12:03:59 +02:00
lissyx
3e60413f27
Merge pull request #2305 from rhamnett/patch-3 
Update import_voxforge.py
 2019-08-18 13:53:09 +02:00
Richard Hamnett
57156fffd0
Update import_voxforge.py 
Fix importer
 2019-08-18 09:44:12 +01:00
Reuben Morais
7f642eda94
Merge pull request #2283 from rcgale/master 
Checking for empty transcripts during character encoding
 2019-08-16 09:18:22 +02:00
Reuben Morais
29836c9dcc
Merge pull request #2289 from mozilla/error-non-finite-loss 
Error message if a sample has non-finite loss
 2019-08-12 10:48:59 +02:00
Robert Gale
85e25fa2d7 Applying text_to_char_array to each row in DataFrame so we can provide wav_filename context on exception 2019-08-07 14:43:41 -07:00
Reuben Morais
86fff2f660
Merge pull request #2265 from mozilla/cudnnrnn_compatible 
Allow loading a CuDNN RNN checkpoint in a CPU-capable graph (Fixes #2264)
 2019-08-07 14:57:16 +02:00
Reuben Morais
248c01001e Error message if a sample has non-finite loss 2019-08-07 10:31:15 +02:00
Reuben Morais
c76070be19
Merge pull request #2282 from mozilla/dynamic-batch-size-in-train-val-graph 
Use dynamic batch size in train/val graph
 2019-08-07 10:03:53 +02:00
dabinat
1cf2d6a8e6
Merge pull request #2287 from dabinat/label-validation 
Label validation - Replace hyphens with spaces
 2019-08-06 06:19:45 -07:00
dabinat
abc05b4a4d Label validation - Replace hyphens with spaces 2019-08-05 09:20:13 -07:00
Robert Gale
a3e0e9f9bc
Update text.py 
"characters" was a bad variable name now that I think about it
 2019-08-01 12:14:13 -07:00
Robert Gale
8ec6ac8079 Checking for empty transcripts during character encoding 
This way we can get a plain English exception early, rather than a matrix shape error during training.
 2019-08-01 11:19:21 -07:00
Reuben Morais
3636d9b481 Use dynamic batch size in train/val graph 
Avoid needing to use the same batch size for training and validation.
 2019-08-01 14:53:46 +02:00
Tilman Kamp
daa6167829
Merge pull request #2268 from tilmankamp/reportfilename 
Fix #2180 - Added wav_filename to WER report
 2019-07-23 17:19:32 +02:00