文件历史

提交图

117 次代码提交

作者 SHA1 备注 提交日期
Nickolay Shmyrev 1496b597d3 Fixes issues with models without alignment 2022-07-03 13:00:05 +02:00
Nickolay Shmyrev 02c40ea612 Bring back incremental decoder, now works better with kaldi fix 2022-05-15 23:20:19 +02:00
Nickolay Shmyrev d5ca98a982 Revert incremental decoder patch, leads to very slow decoding in some cases due to lattice blow-up 2022-05-06 00:34:50 +02:00
Nickolay Shmyrev 5aaea8fc90 Simplify makefile 2022-04-29 02:35:17 +02:00
Chupligin Sergey 5d7752b657 Allow build with shared librares (#939) 2022-04-29 03:23:22 +03:00
Nickolay Shmyrev 7b7d814484 Introduce incremental decoder with confidences in partial results 2022-04-07 01:07:47 +02:00
Nickolay Shmyrev a9bf929ebd Add NLSML output for GPU recognizer 2022-03-03 23:01:49 +01:00
Nickolay Shmyrev 3336cd704b Add resampler 2022-03-03 21:48:25 +01:00
Nickolay Shmyrev a57a84f90e Refactor GPU API to hide the ID and keep it closer to CPU recognizer 2022-03-03 21:09:09 +01:00
Nickolay Shmyrev ad546a8f1a Also read processing options 2022-02-17 10:43:05 +01:00
Nickolay Shmyrev 1f447a8dfc Rename according to Kaldi changes 2022-02-10 20:52:55 +01:00
Nickolay Shmyrev f574d896e9 Emtpy result should be also xml 2022-02-03 23:43:00 +01:00
Nickolay Shmyrev a561c2d6d4 Don't add space before string 2022-02-03 23:26:53 +01:00
Nickolay Shmyrev 79b8395be0 Add NLSML output 2022-02-03 23:08:09 +01:00
Nickolay Shmyrev 2135223490 Put stream information in a single structure 2022-01-12 14:58:43 +01:00
Nickolay Shmyrev 6f86944a06 Implement wave chunking for cuda decoder 2022-01-12 01:32:20 +01:00
Nickolay Shmyrev 9861be2787 Add libs as dependencies in Makefile 2022-01-10 20:21:24 +01:00
Nickolay Shmyrev c6fab363e6 Don't close channel which not yet started 2022-01-09 15:15:20 +01:00
Nickolay Shmyrev 5428d36d16 Round times 2021-12-26 01:12:18 +01:00
Nickolay Shmyrev 525b722c44 Compile without CUDA too 2021-12-24 01:35:06 +01:00
Nickolay Shmyrev 72bf210164 Put the demo into main folder 2021-12-24 01:07:38 +01:00
Nickolay Shmyrev 93e81c3bc8 Bigger frames per chunk for our big models 2021-12-24 00:22:42 +01:00
Nickolay Shmyrev cb0f8e6411 Per-stream wait API 2021-12-23 22:34:47 +01:00
Nickolay Shmyrev 848b2dc753 Expose results in Python 2021-12-17 22:57:00 +01:00
Nickolay Shmyrev 60f0396fe0 Reset lattice on endpoint 2021-12-17 01:13:09 +01:00
Nickolay Shmyrev 344e137a61 Decoding works, results are empty yet 2021-12-13 01:21:59 +01:00
Nickolay Shmyrev 6977be7fb7 Batch recognizer draft 2021-12-12 21:37:44 +01:00
Nickolay Shmyrev ad5bec114d Fix issue with empty lattice 2021-11-08 13:46:35 +01:00
Nickolay Shmyrev 81f58667ff Update Android build 2021-11-07 13:45:27 +01:00
sskorol 13993db542 Added missing cuda lib to Vosk compilation flags 2021-11-06 17:24:49 +02:00
Nickolay Shmyrev 5bcdf454ec Add fbank feature support 2021-10-30 22:45:01 +02:00
Nickolay Shmyrev 4ccdda44ac Extend documentation 2021-10-12 22:35:31 +02:00
Nickolay Shmyrev 5e46825474 Add try/catch wrapper for C++ method to raise native exceptions. Python and Java
are implemented, others on the way
2021-10-12 22:31:36 +02:00
Nickolay Shmyrev 9a3906831b Properly convert lattice. Fixes issue #713 2021-10-08 21:11:00 +02:00
Nickolay Shmyrev 332553ec1e Revert rescoring to more accurate pure fst composition 2021-10-04 01:05:04 +02:00
Nickolay Shmyrev 646af3f652 Explain more about sample rate in constructor 2021-09-17 11:17:38 +02:00
Nickolay Shmyrev e2af710369 Rework rescoring for faster and more accurate results 2021-08-31 21:56:21 +02:00
Nickolay Shmyrev abff8a4f56 Organize makefile structure 2021-07-29 11:58:31 +02:00
Timur 188575f3a2 Added MKL build options (#647)
Co-authored-by: Kasimov Timur <t.kasimov@cft.ru>
2021-07-29 11:03:39 +03:00
Nickolay Shmyrev 5965ea30e6 Fetch_sub returns original value 2021-07-06 12:11:12 +02:00
Nickolay Shmyrev 5c75495c03 Atomic refcounting for thread safety. See
https://github.com/alphacep/vosk-api/issues/606

for more details
2021-07-06 11:09:34 +02:00
Nickolay Shmyrev 942e15484a Protect against multiple resets 2021-06-28 01:59:43 +02:00
Nickolay Shmyrev 558b4dd69e Don't align lattice twice 2021-06-24 20:08:54 +02:00
Nickolay Shmyrev 7f894f784a Better check for the model files 2021-06-21 00:19:55 +02:00
Nickolay Shmyrev 75bedfe06d Add a method to show/hide words and their times 2021-06-07 01:04:37 +02:00
Nickolay Shmyrev 0930def9ec Fix critical accuracy issues, actually use rescoring 2021-05-30 21:58:06 +02:00
Nickolay Shmyrev 499b2f183a Introduce new API to set speaker model to already initialized recognizer.
Introduce a method to reset recognizer results to start from scratch without
computation of the result.
2021-05-26 00:46:32 +02:00
Nickolay Shmyrev f8189685e5 Add max alternatives output 2021-05-19 18:47:25 +02:00
Nickolay Shmyrev ee8663e6d9 RNNLM rescoring support 2021-05-05 19:34:30 +02:00
Nickolay Shmyrev f351280f49 Allow extra cflags 2021-03-30 19:24:00 +02:00