Alexandre Lissy
5d842c284b
Extract TensorFlow version ourselves
2019-01-23 12:51:05 +01:00
Alexandre Lissy
6323a1428f
Proper DeepSpeech error codes
...
Fixes #1782
2019-01-22 19:05:43 +01:00
Alexandre Lissy
0493b9aae5
Fix incorrect previous_state_h identification
...
Fixes #1843
2019-01-22 16:07:17 +01:00
Alexandre Lissy
9b845533c9
Proper TFLite error checking
...
Fixes #1842
2019-01-22 15:53:36 +01:00
Alexandre Lissy
8226dbc6ca
Pull tensors from TFLite model by name
...
Fixes #1835
2019-01-16 23:28:34 +01:00
Reuben Morais
30cf4a3a54
Merge pull request #1834 from mozilla/issue1828
...
Update decoder parameter names in native client (Fixes #1828 )
2019-01-15 14:02:44 +00:00
Alexandre Lissy
08c9c9ab79
Add basic Android native_client testing
...
Fixes #1785
2019-01-15 13:23:49 +01:00
Reuben Morais
366a82d7e7
Update decoder parameter names in native client
2019-01-15 09:32:31 -02:00
Carlos Fonseca M
17517cb453
Add math defines Windows
2018-12-15 11:43:17 -06:00
Reuben Morais
440893c58d
Simplify ctcdecode API signatures to avoid nested STL structures
...
Flatten structure to avoid nested STL structures which are awkward to
wrap with SWIG and slower at runtime.
2018-11-03 07:48:02 -03:00
Reuben Morais
3cc9b3711d
Use ctcdecode in native client
2018-10-25 17:01:08 -03:00
Alexandre Lissy
3dad37eb05
Remove AOT
2018-09-20 17:02:36 +02:00
Reuben Morais
799baf1f99
Fix parameter inference from model shape
2018-09-17 11:20:40 -03:00
Reuben Morais
6f27928841
Free StreamingState pointer with delete instead of new, expose cheap state deallocation API
2018-08-10 12:32:32 -03:00
Reuben Morais
51cac1031b
Address review comments
2018-08-03 14:46:05 -03:00
Reuben Morais
385bc7a65f
Remove word count bonus in favor of length normalization
2018-08-02 13:22:24 -03:00
Reuben Morais
8f810ee001
Add intermediateDecode method to the API
2018-08-02 13:22:24 -03:00
Reuben Morais
9b8871cd94
Infer n_steps and n_context from model file
2018-08-02 13:22:24 -03:00
Reuben Morais
ae9c62cae9
Convert native API to C instead of C++
2018-08-02 13:22:24 -03:00
Reuben Morais
400c9c0e5d
Streaming inference
2018-08-02 13:22:24 -03:00
Alexandre Lissy
6c4d4f1e68
Expose versions numbers on stderr
...
Fixes #1292
2018-03-16 09:46:31 +01:00
Alexandre Lissy
0c323e8c67
Support MemmappedFileSystem
...
Fixes #1231
2018-02-15 12:18:52 +01:00
Alexandre Lissy
bcfd4fd409
Update for r1.5
2018-02-14 11:58:48 +01:00
Alexandre Lissy
ade66b015d
Build TensorFlow as monolithic into libdeepspeech.so
...
Also limit the set of included Ops to avoid wasting size.
2018-01-31 16:53:57 +01:00
Alexandre Lissy
db09ab7d19
Preliminary support for libdeepspeech_model.so
2017-11-03 17:25:25 +01:00
Reuben Morais
ad45ab47e6
Detect and warn about alphabets that don't match the model shape
2017-11-01 19:58:57 -02:00
Reuben Morais
17966809b7
Print error messages in Model construction instead of silently continuing
2017-11-01 19:58:57 -02:00
Reuben Morais
980c66b29c
Add a proper name for the logits output in the graph and use it
2017-10-08 23:32:17 -03:00
Reuben Morais
d8b059c623
Make the new decoder optional
2017-10-08 23:32:11 -03:00
Reuben Morais
02a75041ae
Use the decoder with LM in the native clients
2017-09-29 17:42:07 -03:00
Reuben Morais
1c4cbf1813
Support custom alphabet mappings ( Fixes #692 ) ( #797 )
...
Support custom alphabet mappings
2017-08-31 11:51:15 +02:00
Chris Lord
d829eb4436
Split audioToInputVector into a separate library
...
Split out functions that don't depend on TensorFlow into a separate
library to avoid opening the TensorFlow library if it's not necessary.
2017-06-27 10:16:54 +01:00
Chris Lord
08deda2b92
Improve build process and instructions to fix performance issues
...
Certain optimisations that TensorFlow recommends cause significant
performance penalties in c_speech_features. Explicitly disable these
features in the BUILD file only for c_speech_features.
The build has been restructured to produce a single deepspeech library,
rather than 3 separate deepspeech, c_speech_features and kissfft
libaries. This should make the deepspeech client and libraries
significantly easier to use and distribute.
2017-05-17 10:52:14 +01:00
Chris Lord
616d77ef2d
Various renames in libdeepspeech to allow for clearer more efficient use
...
libdeepspeech now uses the namespace DeepSpeech. The 'DeepSpeech' class
has been renamed to 'Model', 'getMfccFrames' has been renamed to
'getInputVector' and a function outside of the Model class has been added,
'audioToInputVector' to allow it to be used without initialising
TensorFlow and loading a model.
2017-05-17 10:52:14 +01:00
Chris Lord
10068fc40e
Add libdeepspeech python bindings
2017-05-02 10:41:59 +01:00
Chris Lord
c9cd4ff6f6
Convert libdeepspeech into a proper C++ library to ease use/bindings
...
Formerly, libdeepspeech was basically a C library. Rather than returning
a context struct and having every function require this context struct,
instead provide a DeepSpeech class that handles its own destruction.
2017-04-25 18:43:26 +01:00
Chris Lord
8234b3dc56
Update c_speech_features
...
Update brings performance/accuracy fixes and memory leak fixes.
2017-04-12 16:55:52 +01:00
Alexandre Lissy
b726c5c91a
Pass correct input_lengths value as a Tensor to Session::Run()
...
Fixes #473
2017-03-29 11:39:23 +02:00
Reuben Morais
07620c4779
Export graph should take sequence lengths as input
2017-03-28 16:40:35 +02:00
Chris Lord
2740336d08
Implement a native client using TensorFlow C API
2017-02-23 16:35:44 +00:00
