Commit Graph

  • 5f1de523c4 WIP: Add documentation. Fangjun Kuang 2021-08-23 20:22:24 +08:00
  • 894be068e7 Update prepare.sh to create LM training data; add missed scripts local/prepare_lm_training_data.py Daniel Povey 2021-08-23 19:51:58 +08:00
  • 13200d707b Merge remote-tracking branch 'upstream/master' Daniel Povey 2021-08-23 19:13:15 +08:00
  • 26b5b5ba46 Get tests to work for MaskedLmConformer Daniel Povey 2021-08-23 19:05:31 +08:00
  • 5fecd24664 Test, and fix, TransformerDecoderRelPos Daniel Povey 2021-08-23 17:48:00 +08:00
  • 7856ab89fc Test, and fix, TransformerDecoderLayerRelPos Daniel Povey 2021-08-23 17:39:37 +08:00
  • 556fae586f Add testing for MaskedLmConformerEncoder Daniel Povey 2021-08-23 17:22:03 +08:00
  • 2fbe3b78fd Add more testing; fix issue about channel dim of LayerNorm. Daniel Povey 2021-08-23 17:18:00 +08:00
  • 16a420ec8e Begin to add documentation. Fangjun Kuang 2021-08-23 16:09:58 +08:00
  • 57cb611665
    [yesno] Remove padding in TDNN (#21) Fangjun Kuang 2021-08-23 15:59:36 +08:00
  • ece74b7542 Remove padding in the model to make the results reproducible. Fangjun Kuang 2021-08-23 15:48:01 +08:00
  • e0b04ba54f Progress in testing Daniel Povey 2021-08-23 15:38:37 +08:00
  • 2e37b29e66 Disable SpecAug for yesno. Fangjun Kuang 2021-08-23 13:57:46 +08:00
  • 4d849cfd03 More style issue fixes. Fangjun Kuang 2021-08-23 13:28:30 +08:00
  • 7b267e8be6 Fix style issues. Fangjun Kuang 2021-08-23 12:55:02 +08:00
  • 27a0c80af8 Add phone based LF-MMI training. Fangjun Kuang 2021-08-23 12:52:13 +08:00
  • 6c2c9b9d74
    Add recipe for the yes_no dataset. (#16) Fangjun Kuang 2021-08-23 11:36:29 +08:00
  • 03ff4aab2f Some progress on refactoring conformer code, it's in transformer.py only... Daniel Povey 2021-08-23 11:11:09 +08:00
  • c6e3e10ac7 Fix style issues. Fangjun Kuang 2021-08-23 10:56:53 +08:00
  • b06f4cb513 Merge remote-tracking branch 'dan/master' into yesno Fangjun Kuang 2021-08-23 10:44:39 +08:00
  • 19c4214958
    Fix code style and add copyright. (#18) pkufool 2021-08-23 10:43:59 +08:00
  • 90ea10acb0 Update k2 version, remove lhotse from test workflow pkufool 2021-08-23 10:31:46 +08:00
  • 62eef66f5b Install icefall requirements pkufool 2021-08-23 10:18:51 +08:00
  • 2300a15839 Fix lhotse installation pkufool 2021-08-23 10:09:04 +08:00
  • c97f6f63a8 Fix github workflows pkufool 2021-08-23 09:54:38 +08:00
  • f2a9e69223 Reformat code style with black. pkufool 2021-08-23 09:32:38 +08:00
  • 8c75c0abeb Reformat conformer.py by black pkufool 2021-08-23 09:25:17 +08:00
  • 7edc0c6d0a Minor fixes. Fangjun Kuang 2021-08-23 08:55:46 +08:00
  • 22dc936b69 Minor fixes. Fangjun Kuang 2021-08-23 08:37:44 +08:00
  • 6617d5828d Train more epochs for GitHub actions. Fangjun Kuang 2021-08-23 08:30:39 +08:00
  • 3ffcd95086 Minor fixes. Fangjun Kuang 2021-08-23 07:56:06 +08:00
  • 1bdfcb62b9 Fix a typo. Fangjun Kuang 2021-08-23 07:52:33 +08:00
  • f65525d0a2 Add GitHub actions to run yesno. Fangjun Kuang 2021-08-23 07:50:18 +08:00
  • 88166c598b Add Colab notebook for the yesno dataset. Fangjun Kuang 2021-08-22 23:39:43 +08:00
  • 9808d30282 Remove duplicate lines pkufool 2021-08-22 22:19:04 +08:00
  • 09587d1108 Refactoring: Remove unused code. Fangjun Kuang 2021-08-22 22:13:13 +08:00
  • 4e89a43442 Minor fix pkufool 2021-08-22 22:10:22 +08:00
  • b4fd6338bb Fix style and add copyright pkufool 2021-08-22 22:06:28 +08:00
  • 40109c0d93 Add embedding scale to nn.Embedding. Fangjun Kuang 2021-08-22 14:45:39 +08:00
  • 24d3a98378 Merge remote-tracking branch 'upstream/master' Daniel Povey 2021-08-22 11:56:45 +08:00
  • ea43b49ef2 Remove BatchNorm, use LayerNorm Daniel Povey 2021-08-22 11:56:22 +08:00
  • 076a70b62d Initial conformer refactoring, not nearly done Daniel Povey 2021-08-22 11:47:26 +08:00
  • cbe5ee1111 Copy some files, will edit.. Daniel Povey 2021-08-21 22:35:43 +08:00
  • 421a41027a Get dataset.py working.. Daniel Povey 2021-08-21 18:23:46 +08:00
  • f246f0c24b Add recipe for the yes_no dataset. Fangjun Kuang 2021-08-21 17:20:31 +08:00
  • 8469f9ae0a
    Refactor asr_datamodule. (#15) Fangjun Kuang 2021-08-21 09:53:46 +08:00
  • ed16585c58 Minor fixes. Fangjun Kuang 2021-08-21 08:25:34 +08:00
  • 8a8bf67faf Fixes after review. Fangjun Kuang 2021-08-21 08:15:40 +08:00
  • dbc76dbd85 WIP: Refactor asr_datamodule. Fangjun Kuang 2021-08-20 23:44:22 +08:00
  • 0b656e4e1c
    Add a link to Colab. (#14) Fangjun Kuang 2021-08-20 15:43:25 +08:00
  • 2a021f864b Add a link to Colab. Fangjun Kuang 2021-08-20 15:31:51 +08:00
  • 9d0cc9d829
    Support computing nbest oracle WER. (#10) Fangjun Kuang 2021-08-20 11:53:37 +08:00
  • acefc70322 Add usage example with a provided pretrained model. Fangjun Kuang 2021-08-20 11:24:52 +08:00
  • 60211ce12a Merge remote-tracking branch 'dan/master' into nbest-oracle Fangjun Kuang 2021-08-20 10:27:15 +08:00
  • ef233486ae
    The training script produce WER of 2.57% on librispeech test-clean (#13) pkufool 2021-08-20 10:08:08 +08:00
  • a33852fd7a Add RESULTS.md pkufool 2021-08-20 09:53:55 +08:00
  • 3dadffd2b6 Replace scale with lattice-score-scale. Fangjun Kuang 2021-08-19 18:07:17 +08:00
  • d2ae1ba060 Fix conflicts pkufool 2021-08-19 17:38:19 +08:00
  • 3060c5a556 Add grad_clip and weight-decay, small fix of dataloader and masking pkufool 2021-08-19 17:34:28 +08:00
  • f841581fff Merge remote-tracking branch 'dan/master' into nbest-oracle Fangjun Kuang 2021-08-19 16:26:23 +08:00
  • fb1d284116 Minor fixes. Fangjun Kuang 2021-08-19 16:22:09 +08:00
  • eae1674ffa Support decoding with LM rescoring and attention-decoder rescoring. Fangjun Kuang 2021-08-19 16:10:38 +08:00
  • 92a475941b
    Merge 58eb49821916ca87338c2cc69887ac5621c01cc2 into caa0b9e9425af27e0c6211048acb55a76ed5d315 Fangjun Kuang 2021-08-19 15:39:13 +08:00
  • caa0b9e942
    Fix an error in displaying decoding process. (#12) Fangjun Kuang 2021-08-19 14:54:01 +08:00
  • a87a39da8c Fix an error in displaying decoding process. Fangjun Kuang 2021-08-19 14:52:01 +08:00
  • a73d3ed917 Support decoding multiple files at the same time. Fangjun Kuang 2021-08-18 21:20:42 +08:00
  • f731996abe Use torchaudio to extract features. Fangjun Kuang 2021-08-18 19:31:06 +08:00
  • 0fa4875a9a Add script to run pretrained models. Fangjun Kuang 2021-08-18 19:06:41 +08:00
  • 38d06049de Add scale to all nbest based decoding/rescoring methods. Fangjun Kuang 2021-08-18 18:42:30 +08:00
  • 27c46b66ee Add multi round nbest rescoer pkufool 2021-08-18 15:00:13 +08:00
  • 401c1c5143 Support computing nbest oracle WER. Fangjun Kuang 2021-08-18 12:54:01 +08:00
  • 1c3b13c7eb
    Minor fixes. (#9) Fangjun Kuang 2021-08-16 19:01:25 +08:00
  • 9c2e378476 Minor fixes. Fangjun Kuang 2021-08-16 17:39:31 +08:00
  • 56319b0903 Minor fixes. Fangjun Kuang 2021-08-16 17:03:05 +08:00
  • 58eb498219 Set the initial learning rate directly. Fangjun Kuang 2021-08-16 15:35:00 +08:00
  • 02e409b6ce Replace warmup with lr scheduler. Fangjun Kuang 2021-08-16 00:00:53 +08:00
  • 0be42bef69 Replace warmup with lr scheduler. Fangjun Kuang 2021-08-15 22:59:51 +08:00
  • 21292066ec Fix OOM handling when using DDP. Fangjun Kuang 2021-08-15 18:49:12 +08:00
  • 14e0886559 Minor fixes. Fangjun Kuang 2021-08-15 11:45:53 +08:00
  • 72c0220830 Fix oom handling. Fangjun Kuang 2021-08-15 09:52:17 +08:00
  • 36ac512d00 Add madam optimizer from Dan. Fangjun Kuang 2021-08-14 23:03:50 +08:00
  • c26eb679a5 Merge remote-tracking branch 'dan/master' into doc Fangjun Kuang 2021-08-14 22:58:45 +08:00
  • 12a2fd023e
    Add doc about installation and usage (#7) Fangjun Kuang 2021-08-12 12:44:04 +08:00
  • f0ee6cf0dc Minor fixes after review. Fangjun Kuang 2021-08-12 10:33:50 +08:00
  • b7133f30bd fix typos Fangjun Kuang 2021-08-10 20:20:30 +08:00
  • dec6ecf4da Add TOC. Fangjun Kuang 2021-08-10 20:19:03 +08:00
  • 55be10534d Add readme. Fangjun Kuang 2021-08-10 20:08:23 +08:00
  • 0669aa8ab9 Add attention rescore pipeline pkufool 2021-08-09 12:47:11 +08:00
  • 03242b3328 Remove unused files. Fangjun Kuang 2021-08-07 18:10:41 +08:00
  • 897307f445 Add MMI training with word pieces. Fangjun Kuang 2021-08-07 16:41:16 +08:00
  • 286dce7b0f Merge branch 'master' into nbest pkufool 2021-08-04 15:52:14 +08:00
  • f03c991781 Merge remote-tracking branch 'dan/master' into mmi Fangjun Kuang 2021-08-04 15:00:21 +08:00
  • b1b21eb1e4 Fix decoder padding mask. Fangjun Kuang 2021-08-04 14:57:06 +08:00
  • 5a0b9bcb23
    Refactoring (#4) Fangjun Kuang 2021-08-04 14:53:02 +08:00
  • cabe8b625b Copy the files related to multi round nbest rescoring from k2 & snowfall pkufool 2021-08-04 14:27:11 +08:00
  • a6d9b3c9ab Minor fixes. Fangjun Kuang 2021-08-03 22:16:34 +08:00
  • 2be7a0a555 Remove unused code. Fangjun Kuang 2021-08-03 17:24:06 +08:00
  • f6091b10c0 Refactor transformer.py Fangjun Kuang 2021-08-02 23:48:26 +08:00
  • 1fa30998da WIP: Refactoring Fangjun Kuang 2021-07-31 20:24:47 +08:00
  • c72a11ea1f Merge remote-tracking branch 'dan/master' into style-check Fangjun Kuang 2021-07-31 16:49:54 +08:00