Commit Graph

  • 178b19f166 fix flake8 conflict PingFeng Luo 2021-12-31 19:27:00 +08:00
  • bcc117879a fix flake8 conflict PingFeng Luo 2021-12-31 19:24:35 +08:00
  • 54155cc164 fix style PingFeng Luo 2021-12-31 19:19:42 +08:00
  • 4e92553066 fix sytle PingFeng Luo 2021-12-31 19:14:53 +08:00
  • 6813b754e9 fix style PingFeng Luo 2021-12-31 19:11:24 +08:00
  • 46eab8b727 fix sytle PingFeng Luo 2021-12-31 18:57:06 +08:00
  • 28d1e8660e fix sytle PingFeng Luo 2021-12-31 18:50:12 +08:00
  • 503275e649 Merge branch 'master' into wenetspeech PingFeng Luo 2021-12-31 18:28:22 +08:00
  • f68dc1893f add transducer-stateless-aishell recipe to readme PingFeng Luo 2021-12-31 18:24:26 +08:00
  • b49510e2bf Add label smoothing for transducer loss. Fangjun Kuang 2021-12-31 15:52:33 +08:00
  • 6bb949eb5a extract middle layer memory embedding Guo Liyong 2021-12-31 15:38:53 +08:00
  • a94f7b440e typo PingFeng Luo 2021-12-31 11:41:39 +08:00
  • 4166f5706e add transducer_stateless with char unit to AIShell PingFeng Luo 2021-12-31 09:54:24 +08:00
  • 1717b26cab fix style PingFeng Luo 2021-12-31 09:39:39 +08:00
  • a82d826987 fix style PingFeng Luo 2021-12-30 18:56:04 +08:00
  • 8b8cf6c68e fix style PingFeng Luo 2021-12-30 18:43:47 +08:00
  • f49de8e8e0 Merge branch 'master' of github.com:pingfengluo/icefall PingFeng Luo 2021-12-30 12:08:30 +08:00
  • 18cdea4745 add transducer_stateless with char unit to AIShell PingFeng Luo 2021-12-29 15:34:50 +08:00
  • 7391f4febf Add explaination for some functions and classes Mingshuang Luo 2021-12-30 10:24:47 +08:00
  • 6e5b189fc5 DynamicBucketingSampler wgb14 2021-12-29 15:22:46 -05:00
  • 7828c6ff73 Minor fixes. Fangjun Kuang 2021-12-30 00:15:23 +08:00
  • 283bd126c5 add pretrained.py Mingshuang Luo 2021-12-29 19:10:56 +08:00
  • 8541dc73f9 WIP: Use optimized_transducer to compute transducer loss. Fangjun Kuang 2021-12-28 20:11:01 +08:00
  • 413b2e8569
    Add git sha1 to RESULTS.md for conformer encoder + stateless decoder. (#160) Fangjun Kuang 2021-12-28 12:04:01 +08:00
  • 592b7d9edd Add git sha1 to RESULTS.md for conformer encoder + stateless decoder. Fangjun Kuang 2021-12-28 12:01:05 +08:00
  • 3c89734b79 Use similar number of parameters as conformer encoder. Fangjun Kuang 2021-12-28 11:09:43 +08:00
  • ec083e93d8 Use a stateless decoder. Fangjun Kuang 2021-12-28 10:49:50 +08:00
  • 2cf1b56cb3 Remove SOS from decoder. Fangjun Kuang 2021-12-28 10:38:22 +08:00
  • 65212ee004 Remove grad clipping as there are no RNNs in the network. Fangjun Kuang 2021-12-28 10:31:20 +08:00
  • 2b5881f4d9 Update README.md Mingshuang Luo 2021-12-27 21:18:49 +08:00
  • 514cd1fcf3 Update README.md Mingshuang Luo 2021-12-27 17:01:31 +08:00
  • 7b3624006f update the codes Mingshuang Luo 2021-12-27 16:56:15 +08:00
  • 234307f33a Merge branch 'master' of https://github.com/k2-fsa/icefall PingFeng Luo 2021-12-27 16:43:39 +08:00
  • 14c93add50
    Remove batchnorm, weight decay, and SOS from transducer conformer encoder (#155) Fangjun Kuang 2021-12-27 16:01:10 +08:00
  • 8eea3414a4 Update results. Fangjun Kuang 2021-12-27 15:31:35 +08:00
  • b5735ae16f Merge remote-tracking branch 'dan/master' into rnnt-stateless2 Fangjun Kuang 2021-12-27 15:07:01 +08:00
  • 8187d6236c
    Minor fix to maximum number of symbols per frame for RNN-T decoding. (#157) Fangjun Kuang 2021-12-24 21:48:40 +08:00
  • b2d43c3551
    Merge branch 'k2-fsa:master' into grid-a-vsr-recipe Mingshuang Luo 2021-12-24 14:34:15 +08:00
  • 149ccd1b85 update lipnet-ctc-vsr for grid Mingshuang Luo 2021-12-24 14:16:12 +08:00
  • c57798661c Make --context-size configurable. Fangjun Kuang 2021-12-24 14:12:32 +08:00
  • f033379805 update lipnet-ctc-vsr Mingshuang Luo 2021-12-24 14:07:57 +08:00
  • 60696d3eb2 Minor fixes. Fangjun Kuang 2021-12-24 11:16:29 +08:00
  • 0fa4ca7f02 Minor fix to maximum number of symbols per frame RNN-T decoding. Fangjun Kuang 2021-12-24 11:05:47 +08:00
  • eb29202168 Minor fixes pkufool 2021-12-24 08:05:41 +08:00
  • 2d6631ac76 Replace torchaudio rnnt_loss to k2 pruned rnnt loss pkufool 2021-12-24 07:56:42 +08:00
  • e31c14b335 add reference Guo Liyong 2021-12-23 19:37:43 +08:00
  • a9ad9553b5 use wav2vec as a teacher model Guo Liyong 2021-12-23 19:08:35 +08:00
  • 3b42f0347f train with full libri Guo Liyong 2021-12-23 18:39:13 +08:00
  • 8985440ce1 copy quantization files from dan's repo Guo Liyong 2021-12-23 18:33:39 +08:00
  • 3570cb738a test exclude Guo Liyong 2021-12-23 18:31:13 +08:00
  • c59bdf651a
    Merge 54bcc167e17e0edb26f9d32a1f35510fbf323fee into 5b6699a8354b70b23b252b371c612a35ed186ec2 LIyong.Guo 2021-12-23 18:07:27 +08:00
  • c7df25bd5b add transducer_stateless to AIShell PingFeng Luo 2021-12-23 14:59:35 +08:00
  • 35d63de820 Remove batchnorm, weight decay, and SOS. Fangjun Kuang 2021-12-23 14:19:49 +08:00
  • 5b6699a835
    Minor fixes to the RNN-T Conformer model (#152) Fangjun Kuang 2021-12-23 13:54:25 +08:00
  • 225c8a03b6 Fix typos. Fangjun Kuang 2021-12-23 12:56:09 +08:00
  • ca0d7c5795 Minor fixes. Fangjun Kuang 2021-12-23 12:35:25 +08:00
  • 1b45d37080 Merge branch 'master' of https://github.com/k2-fsa/icefall PingFeng Luo 2021-12-23 11:24:33 +08:00
  • 9a62a0e7bc Merge remote-tracking branch 'dan/master' into rnnt-fix-joiner Fangjun Kuang 2021-12-23 10:36:45 +08:00
  • fb6a57e9e0
    Increase the size of the context in the RNN-T decoder. (#153) Fangjun Kuang 2021-12-23 07:55:02 +08:00
  • 834114f1c2
    Update decoder.py Fangjun Kuang 2021-12-22 23:49:31 +08:00
  • c24b610389 Minor fixes. Fangjun Kuang 2021-12-22 23:47:24 +08:00
  • 32d96002d9 Run pretrained transducer_stateless model in CI. Fangjun Kuang 2021-12-22 23:32:39 +08:00
  • 135c797948 Add link to colab notebook. Fangjun Kuang 2021-12-22 23:28:22 +08:00
  • 0a15bee545 Update README and RESULTS. Fangjun Kuang 2021-12-22 22:56:28 +08:00
  • 8a3c2a00db Add pretrained.py and export.py for transducer_stateless. Fangjun Kuang 2021-12-22 19:30:14 +08:00
  • 1abf255bdd update some files Mingshuang Luo 2021-12-22 16:11:35 +08:00
  • 61915876c5 Fix an error after refactoring. Fangjun Kuang 2021-12-22 15:41:17 +08:00
  • 7f4749a4d6 Fix CI errors. Fangjun Kuang 2021-12-22 15:26:26 +08:00
  • fbc1bc3a6b Implement beam search. Fangjun Kuang 2021-12-22 15:15:58 +08:00
  • a9abcc5fda Add grid AVSR task results Mingshuang Luo 2021-12-22 11:20:41 +08:00
  • afec6b6cae Update greedy search for modified decoder. Fangjun Kuang 2021-12-21 11:37:56 +08:00
  • 27bfcc4ea8 Add grid ASR task results Mingshuang Luo 2021-12-20 15:13:41 +08:00
  • 4aa3149084
    Merge branch 'k2-fsa:master' into grid-a-vsr-recipe Mingshuang Luo 2021-12-20 11:06:22 +08:00
  • 04977175a3 Increase the size of the context in the RNN-T decoder. Fangjun Kuang 2021-12-18 23:54:31 +08:00
  • d362a3dba7 Reduce the number of decoder layers from 4 to 2. Fangjun Kuang 2021-12-18 11:25:33 +08:00
  • 9d0d5d19fb Remove sos ID. Fangjun Kuang 2021-12-18 11:22:12 +08:00
  • 63e1266e3a Use tanh in the joint network. Fangjun Kuang 2021-12-18 11:14:05 +08:00
  • 66cc9b4592 Replace BatchNorm in the Conformer model with LayerNorm. Fangjun Kuang 2021-12-18 11:12:38 +08:00
  • 4635af633a Remove input feature batchnorm.. Fangjun Kuang 2021-12-18 11:05:28 +08:00
  • 4eb5e7864a Disable weight decay. Fangjun Kuang 2021-12-18 11:03:10 +08:00
  • cb04c8a750
    Limit the number of symbols per frame in RNN-T decoding. (#151) Fangjun Kuang 2021-12-18 11:00:42 +08:00
  • f8d02d633c Limit the number of symbols per frame in RNN-T decoding. Fangjun Kuang 2021-12-18 10:13:33 +08:00
  • 1d44da845b
    RNN-T Conformer training for LibriSpeech (#143) Fangjun Kuang 2021-12-18 07:42:51 +08:00
  • 9d68199322 Fix tests. Fangjun Kuang 2021-12-17 20:51:50 +08:00
  • 9fad0fd915 Minor fixes. Fangjun Kuang 2021-12-17 20:39:34 +08:00
  • be493ad913 Minor fixes. Fangjun Kuang 2021-12-17 20:22:26 +08:00
  • 270febb638 Fix tests. Fangjun Kuang 2021-12-17 20:21:04 +08:00
  • 9639f6dc0a Minor fixes. Fangjun Kuang 2021-12-17 20:19:36 +08:00
  • 47b0f2ec2f Update RESULT.md to include RNN-T Conformer. Fangjun Kuang 2021-12-17 20:16:47 +08:00
  • d7eb94c4c9 Fix README. Fangjun Kuang 2021-12-17 19:46:28 +08:00
  • 164321c79d Minor fixes to make it ready for merge. Fangjun Kuang 2021-12-17 19:43:04 +08:00
  • f6a33a85c5 Use stateless decoder. Fangjun Kuang 2021-12-17 16:48:57 +08:00
  • bea78f6094 lazy loading and use SingleCutSampler wgb14 2021-12-17 00:38:52 -05:00
  • fcc22d3e91 Use LSTM layers for the encoder. Fangjun Kuang 2021-12-17 11:58:30 +08:00
  • 532309bf72 Add conformer.py without pre-commit checking Guanbo Wang 2021-12-16 20:20:41 -05:00
  • 76a289126f add conformer training recipe wgb14 2021-12-16 20:18:02 -05:00
  • 71ef6a9e11 Merge remote-tracking branch 'upstream/master' into gigaspeech_recipe Guanbo Wang 2021-12-16 19:13:14 -05:00
  • 738eeea301 Merge branch 'grid-a-vsr-recipe' of https://github.com/luomingshuang/icefall into grid-a-vsr-recipe Mingshuang Luo 2021-12-15 23:29:41 +08:00
  • a5c1bcd58c Update prepare.sh Mingshuang Luo 2021-12-15 23:29:37 +08:00
  • e42730d08c
    Merge branch 'k2-fsa:master' into master Mingshuang Luo 2021-12-15 23:18:45 +08:00