mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-18 22:02:24 +00:00
Code Issues Packages Projects Releases Wiki Activity
84 Commits 5 Branches 3 Tags
Commit Graph

71 Commits

Author SHA1 Message Date
Daniel Povey
0cfa8c80e0 Merge remote-tracking branch 'upstream/master' into conformer_lm 2021-09-24 11:21:50 +08:00
Fangjun Kuang
a80e58e15d
Refactor decode.py to make it more readable and more modular. (#44) 
* Refactor decode.py to make it more readable and more modular.

* Fix an error.

Nbest.fsa should always have token IDs as labels and
word IDs as aux_labels.

* Add nbest decoding.

* Compute edit distance with k2.

* Refactor nbest-oracle.

* Add rescore with nbest lists.

* Add whole-lattice rescoring.

* Add rescoring with attention decoder.

* Refactoring.

* Fixes after refactoring.

* Fix a typo.

* Minor fixes.

* Replace [] with () for shapes.

* Use k2 v1.9

* Use Levenshtein graphs/alignment from k2 v1.9

* [doc] Require k2 >= v1.9

* Minor fixes.
 2021-09-20 15:44:54 +08:00
Wei Kang
9a6e0489c8
update api for RaggedTensor (#45) 
* Fix code style

* update k2 version in CI

* fix compile hlg
 2021-09-14 16:39:56 +08:00
Daniel Povey
3ce1de337d UPdates for new k2 version; change LR decay from 0.85 to 0.9 2021-09-13 20:57:02 +08:00
Wei Kang
24656e9749
Update docs and remove unnecessary arguments (#42) 
* Fix typo in docs

* Update docs and remove unnecessary arguments

* Fix code style
 2021-09-13 18:28:57 +08:00
Fangjun Kuang
f792b466bf
Change default value of lattice-score-scale from 1.0 to 0.5 (#41) 
* Change the default value of lattice-score-scale from 1.0 to 0.5

* Fix CI.
 2021-09-13 10:49:18 +08:00
Daniel Povey
d0e5b9b8a5 Change to exp_5, 1/sqrt(t) component. 2021-09-09 14:08:19 +08:00
Fangjun Kuang
7f8e3a673a
Add commands for reproducing. (#40) 
* Add commands for reproducing.

* Use --bucketing-sampler by default.
 2021-09-09 13:50:31 +08:00
Fangjun Kuang
abadc71415
Use new APIs with k2.RaggedTensor (#38) 
* Use new APIs with k2.RaggedTensor

* Fix style issues.

* Update the installation doc, saying it requires at least k2 v1.7

* Use k2 v1.7
 2021-09-08 14:55:30 +08:00
Daniel Povey
56a88badd1 Move to Gloam optimizer, exponential lrate 2021-09-08 13:59:50 +08:00
Daniel Povey
d313c27c14 Change configuration again.. not great performance. 2021-09-07 20:58:00 +08:00
Daniel Povey
573e0582d8 Run in exp_2, with foam from start, knee_factor=5.0, initial_lrate=2e-04. 2021-08-30 14:10:21 +08:00
Daniel Povey
ccf7bdec23 Add Foam optimizer; I used this from epoch 3. 2021-08-28 21:51:54 +08:00
Daniel Povey
d045831a4f Get dataset to work for empty input sentences; test it 2021-08-25 15:54:36 +08:00
Fangjun Kuang
184dbb3ea5
Add documentation about code style and creating new recipes. (#27) 2021-08-25 14:48:41 +08:00
Daniel Povey
a7b61100de Use collate_fn as class. harmless but not necessary without multiple workers 2021-08-25 11:27:47 +08:00
Daniel Povey
0d97e689be Version I am running... 2021-08-24 21:59:41 +08:00
Fangjun Kuang
96e7f5c7ea
Release v0.1 (#26) 2021-08-24 21:30:30 +08:00
pkufool
f4223ee110
Add TDNN-LSTM-CTC Results (#25) 
* Add tdnn-lstm pretrained model and results

* Add docs for TDNN-LSTM-CTC

* Minor fix

* Fix typo

* Fix style checking
 2021-08-24 21:09:27 +08:00
Fangjun Kuang
1bd5dcc8ac
WIP: Add doc for the LibriSpeech recipe. (#24) 
* WIP: Add doc for the LibriSpeech recipe.

* Add more doc for LibriSpeech recipe.

* Add more doc for the LibriSpeech recipe.

* More doc.
 2021-08-24 20:28:32 +08:00
Daniel Povey
e6eefeba88 Changes to dataset to prevent OOM on batches with short sentences 2021-08-24 14:50:49 +08:00
Fangjun Kuang
01da00dca0
WIP: Add documentation. (#22) 
* Begin to add documentation.

* WIP: Add documentation.

* Fix a typo.

* Add more doc for the recipe yesno.

* Add more doc for the yesno recipe.
 2021-08-24 14:28:08 +08:00
Daniel Povey
9576d6574f Various bug fixes 2021-08-23 23:45:03 +08:00
Daniel Povey
7711fba867 Fix bugs; first version that is running successfully. 2021-08-23 22:40:23 +08:00
Daniel Povey
c3a8727446 Add train.py 2021-08-23 22:28:45 +08:00
Daniel Povey
894be068e7 Update prepare.sh to create LM training data; add missed scripts local/prepare_lm_training_data.py 2021-08-23 19:51:58 +08:00
Daniel Povey
13200d707b Merge remote-tracking branch 'upstream/master' 2021-08-23 19:13:15 +08:00
Daniel Povey
26b5b5ba46 Get tests to work for MaskedLmConformer 2021-08-23 19:05:31 +08:00
Daniel Povey
5fecd24664 Test, and fix, TransformerDecoderRelPos 2021-08-23 17:48:00 +08:00
Daniel Povey
7856ab89fc Test, and fix, TransformerDecoderLayerRelPos 2021-08-23 17:39:37 +08:00
Daniel Povey
556fae586f Add testing for MaskedLmConformerEncoder 2021-08-23 17:22:03 +08:00
Daniel Povey
2fbe3b78fd Add more testing; fix issue about channel dim of LayerNorm. 2021-08-23 17:18:00 +08:00
Fangjun Kuang
57cb611665
[yesno] Remove padding in TDNN (#21) 
* Disable SpecAug for yesno.

Also replace Adam with SGD.

* Remove padding in the model to make the results reproducible.
 2021-08-23 15:59:36 +08:00
Daniel Povey
e0b04ba54f Progress in testing 2021-08-23 15:38:37 +08:00
Fangjun Kuang
6c2c9b9d74
Add recipe for the yes_no dataset. (#16) 
* Add recipe for the yes_no dataset.

* Refactoring: Remove unused code.

* Add Colab notebook for the yesno dataset.

* Add GitHub actions to run yesno.

* Fix a typo.

* Minor fixes.

* Train more epochs for GitHub actions.

* Minor fixes.

* Minor fixes.

* Fix style issues.
 2021-08-23 11:36:29 +08:00
Daniel Povey
03ff4aab2f Some progress on refactoring conformer code, it's in transformer.py only... 2021-08-23 11:11:09 +08:00
pkufool
19c4214958
Fix code style and add copyright. (#18) 
* Fix style and add copyright

* Minor fix

* Remove duplicate lines

* Reformat conformer.py by black

* Reformat code style with black.

* Fix github workflows

* Fix lhotse installation

* Install icefall requirements

* Update k2 version, remove lhotse from test workflow
 2021-08-23 10:43:59 +08:00
Daniel Povey
24d3a98378 Merge remote-tracking branch 'upstream/master' 2021-08-22 11:56:45 +08:00
Daniel Povey
ea43b49ef2 Remove BatchNorm, use LayerNorm 2021-08-22 11:56:22 +08:00
Daniel Povey
076a70b62d Initial conformer refactoring, not nearly done 2021-08-22 11:47:26 +08:00
Daniel Povey
cbe5ee1111 Copy some files, will edit.. 2021-08-21 22:35:43 +08:00
Daniel Povey
421a41027a Get dataset.py working.. 2021-08-21 18:23:46 +08:00
Fangjun Kuang
8469f9ae0a
Refactor asr_datamodule. (#15) 
* WIP: Refactor asr_datamodule.

* Fixes after review.

* Minor fixes.
 2021-08-21 09:53:46 +08:00
Fangjun Kuang
0b656e4e1c
Add a link to Colab. (#14) 
It demonstrates the usages of pre-trained models.
 2021-08-20 15:43:25 +08:00
Fangjun Kuang
9d0cc9d829
Support computing nbest oracle WER. (#10) 
* Support computing nbest oracle WER.

* Add scale to all nbest based decoding/rescoring methods.

* Add script to run pretrained models.

* Use torchaudio to extract features.

* Support decoding multiple files at the same time.

Also, use kaldifeat for feature extraction.

* Support decoding with LM rescoring and attention-decoder rescoring.

* Minor fixes.

* Replace scale with lattice-score-scale.

* Add usage example with a provided pretrained model.
 2021-08-20 11:53:37 +08:00
pkufool
ef233486ae
The training script produce WER of 2.57% on librispeech test-clean (#13) 
* Add grad_clip and weight-decay, small fix of dataloader and masking

* Add RESULTS.md
 2021-08-20 10:08:08 +08:00
Fangjun Kuang
caa0b9e942
Fix an error in displaying decoding process. (#12) 2021-08-19 14:54:01 +08:00
Fangjun Kuang
1c3b13c7eb
Minor fixes. (#9) 2021-08-16 19:01:25 +08:00
Fangjun Kuang
12a2fd023e
Add doc about installation and usage (#7) 
* Add readme.

* Add TOC.

* fix typos

* Minor fixes after review.
 2021-08-12 12:44:04 +08:00
Fangjun Kuang
5a0b9bcb23
Refactoring (#4) 
* Fix an error in TDNN-LSTM training.

* WIP: Refactoring

* Refactor transformer.py

* Remove unused code.

* Minor fixes.
 2021-08-04 14:53:02 +08:00