Commit Graph

  • fee72b9cbb add modified-beam-search for streaming decoding luomingshuang 2022-07-25 16:07:26 +08:00
  • 245515fbfe fix causal pkufool 2022-07-25 14:56:53 +08:00
  • 67dbb25620 deploy: d99796898cc369123dfdea8a0f660fe174a33c35 csukuangfj 2022-07-25 04:08:43 +00:00
  • d99796898c
    Update doc to add a link to Nadira Povey's YouTube channel. (#492) Fangjun Kuang 2022-07-25 12:06:40 +08:00
  • dd5ab33239 fix a typo Fangjun Kuang 2022-07-25 12:05:04 +08:00
  • b8e6d39e9b Merge branch 'fisher_swbd' of https://github.com/GoVivace/icefall into fisher_swbd s-mousmita 2022-07-25 00:00:07 -04:00
  • 3fb7393b06 Update to prepare.sh s-mousmita 2022-07-24 23:58:35 -04:00
  • 3c4f8a0357 Update doc to add a link to Nadira Povey's YouTube channel. Fangjun Kuang 2022-07-25 11:48:28 +08:00
  • f4a189a84b Add real streamng ci pkufool 2022-07-25 10:12:04 +08:00
  • fe595f8772 Improve debugging output. Daniel Povey 2022-07-25 09:02:36 +08:00
  • 854c2965a9 Fix bug regarding G_prime being zero Daniel Povey 2022-07-25 06:57:52 +08:00
  • 3acdf3b395 Reworking the computation of Z to be numerically better. Daniel Povey 2022-07-25 06:37:26 +08:00
  • 5513f7fee5 Initial version of fixing numerical issue, will continue though Daniel Povey 2022-07-25 06:27:01 +08:00
  • b0f0c6c4ab Setting lr_update_period=(200,4k) in train.py Daniel Povey 2022-07-25 04:38:12 +08:00
  • 06718052ec Refactoring, putting tunable values in constructor, a little cleanup Daniel Povey 2022-07-25 04:31:42 +08:00
  • 428afe8089
    Merge branch 'k2-fsa:master' into fisher_swbd GoVivace 2022-07-24 15:45:37 -04:00
  • 0adfe6595d modify doc yaozengwei 2022-07-24 20:00:57 +08:00
  • 9576b2cb2c add stats for other 2 recipes yaozengwei 2022-07-24 19:47:30 +08:00
  • 6455fa40b8 Merge remote-tracking branch 'k2-fsa/master' into fix_tensorboard yaozengwei 2022-07-24 19:42:13 +08:00
  • fdb8371c8c add stats for other recipes yaozengwei 2022-07-24 19:41:09 +08:00
  • 8efc512823 Remove some debugging code, found the mismatch Daniel Povey 2022-07-24 11:52:10 +08:00
  • ba96439c76 Saving version I am trying to debug Daniel Povey 2022-07-24 11:00:40 +08:00
  • 962e95f119 Using a more flexible test. Moved to simpler update , tuned diffrently. Daniel Povey 2022-07-24 09:20:53 +08:00
  • b8a9485011 Print git version for test output Daniel Povey 2022-07-24 06:54:29 +08:00
  • 48ac7e0bc3 Add max as well as min to G_prime Daniel Povey 2022-07-24 06:50:05 +08:00
  • 6290fcb535 Cleanup and refactoring Daniel Povey 2022-07-24 05:48:38 +08:00
  • 8a9bbb93bc Cosmetic fixes Daniel Povey 2022-07-24 04:45:57 +08:00
  • 966ac36cde Fixes to comments Daniel Povey 2022-07-24 04:36:41 +08:00
  • 33ffd17515 Some cleanup Daniel Povey 2022-07-24 04:22:11 +08:00
  • 3c4c615e1f Fix comments pkufool 2022-07-23 22:42:18 +08:00
  • e230654aba updating swbd prep part s-mousmita 2022-07-23 10:40:34 -04:00
  • 5393b281ff Adding fisher swbd recipe training with pruned_transducer_stateless2 s-mousmita 2022-07-23 08:48:27 -04:00
  • 353863a55c modified beam search for stateless3,4 pkufool 2022-07-23 17:18:47 +08:00
  • 72d76a4ff8 refactor pkufool 2022-07-23 16:58:43 +08:00
  • ddceb7963b Interpolate between iterative estimate of scale, and original value. Daniel Povey 2022-07-23 15:27:48 +08:00
  • 2c4bdd0ad0 Add _update_param_scales_simple(), add documentation Daniel Povey 2022-07-23 14:49:58 +08:00
  • 9730352257 Redce smoothing constant slightly Daniel Povey 2022-07-23 13:12:31 +08:00
  • e1873fc0bb Tune phase2 again, from 0.005,5.0 to 0.01,40. Epoch 140 is 0.21/0.149 Daniel Povey 2022-07-23 10:10:01 +08:00
  • 0fc58bac56 More tuning, epoch-140 results are 0.23,0.11 Daniel Povey 2022-07-23 09:45:05 +08:00
  • 34a2d331bf Smooth in opposite orientation to G Daniel Povey 2022-07-23 09:38:16 +08:00
  • a972655a70 Tuning. Daniel Povey 2022-07-23 09:15:49 +08:00
  • b47433b77a Fix bug in smooth_cov, for power==1.0 Daniel Povey 2022-07-23 09:06:03 +08:00
  • cc388675a9 Bug fix RE rankj Daniel Povey 2022-07-23 08:24:59 +08:00
  • dee496145d this version performs way worse but has bugs fixed, can optimize from here. Daniel Povey 2022-07-23 08:11:20 +08:00
  • dd10eb140f First version after refactorization and changing the math, where optim.py runs Daniel Povey 2022-07-23 06:32:56 +08:00
  • 1b6daecc63 Add modified_beam_search for pruned_transducer_stateless/streaming_decode.py pkufool 2022-07-22 20:03:43 +08:00
  • 116d0cf26d
    CTC attention model with reworked Conformer encoder and reworked Transformer decoder (#462) Quandwang 2022-07-22 15:31:25 +08:00
  • 9ac0cdd62a
    Update egs/librispeech/ASR/RESULTS.md Quandwang 2022-07-22 15:21:23 +08:00
  • d103106d4c
    Update egs/librispeech/ASR/RESULTS.md Quandwang 2022-07-22 15:20:45 +08:00
  • c1e379973d
    Update egs/librispeech/ASR/RESULTS.md Quandwang 2022-07-22 15:20:34 +08:00
  • d71e35d85e upload the decoding results of rnn-lm to the RESULTS Quandwang 2022-07-22 15:15:53 +08:00
  • 79d7c94397 upload the decoding results of rnn-lm to the RESULTS Quandwang 2022-07-22 14:52:12 +08:00
  • 1b9af23623 Set overwrite=True when extracting features in batches. Fangjun Kuang 2022-07-22 13:48:41 +08:00
  • bd043b0ff0 changes codes for fast_beam_search and export cpu jit luomingshuang 2022-07-22 10:26:19 +08:00
  • 728eb5075f skip scan_pessimistic_batches_for_oom if params.start_batch > 0 Fangjun Kuang 2022-07-22 10:20:36 +08:00
  • efdbb98583 Merge remote-tracking branch 'dan/master' into fix-loading-sampler-state-dict Fangjun Kuang 2022-07-22 10:10:08 +08:00
  • 4da4e69fba Draft of new way of smoothing param_rms, diagonalized by grad Daniel Povey 2022-07-22 06:37:20 +08:00
  • b9fda2cb1c update results Quandwang 2022-07-21 22:06:09 +08:00
  • b354368685 update results Quandwang 2022-07-21 22:00:53 +08:00
  • dd393b5d72 style correction Quandwang 2022-07-21 20:16:56 +08:00
  • 878026864c check whether q k v weight is None Quandwang 2022-07-21 20:12:19 +08:00
  • a1a84ed148 check whether q k v weight is None Quandwang 2022-07-21 20:07:08 +08:00
  • 1f32d34e9b check whether q k v weight is None Quandwang 2022-07-21 20:01:37 +08:00
  • eb86873ee6 minor changes Quandwang 2022-07-21 17:50:39 +08:00
  • 4634911dc2 add rnn lm decoding Quandwang 2022-07-21 17:49:51 +08:00
  • 40eb8c43c9 minor changes to the scripts Quandwang 2022-07-21 15:58:09 +08:00
  • 1ad9862b39 Merge branch 'conformer_ctc2' of github.com:WayneWiser/icefall into conformer_ctc2 Quandwang 2022-07-21 13:35:35 +08:00
  • 713cfc1d79 use original beam size Quandwang 2022-07-21 13:35:23 +08:00
  • ad31021479
    Merge branch 'k2-fsa:master' into conformer_ctc2 Quandwang 2022-07-21 13:24:48 +08:00
  • 6d77f4c239 add streaming decode luomingshuang 2022-07-21 12:01:24 +08:00
  • 3d2986b4c2
    Update conformer.py for aishell4 (#484) Mingshuang Luo 2022-07-20 21:32:53 +08:00
  • 2daf4fec4c add for utt_duration yaozengwei 2022-07-20 20:05:34 +08:00
  • d812f82ea5
    Merge branch 'k2-fsa:master' into conformer_ctc2 Quandwang 2022-07-20 19:47:36 +08:00
  • c88bf2a43d add stats about duration and padding proportion yaozengwei 2022-07-20 18:55:22 +08:00
  • b441dffa83 fix typos and modify the expr of ScaledEmbedding Quandwang 2022-07-20 17:46:40 +08:00
  • 3385fd48b0 add strict=False when model.load_state_dict luomingshuang 2022-07-20 15:51:46 +08:00
  • 9b32670478 update conformer.py luomingshuang 2022-07-20 15:20:38 +08:00
  • c8123b0e37 update conformer.py for aishell4 luomingshuang 2022-07-20 12:33:37 +08:00
  • a8696b36fc
    Merge pull request #483 from yaozengwei/fix_diagnostic Daniel Povey 2022-07-18 23:33:45 -07:00
  • a35b28cd8d fix for case of None stats yaozengwei 2022-07-19 14:29:23 +08:00
  • a63afe348a Increase max_lr_factor from 3.0 to 4.0 Daniel Povey 2022-07-19 06:56:41 +08:00
  • 79a2f09f62 Change how formula for max_lr_factor works, and increase factor from 2.5 to 3. Daniel Povey 2022-07-19 06:54:49 +08:00
  • 525c097130 Increase power from 0.7 to 0.75 Daniel Povey 2022-07-19 05:44:03 +08:00
  • 2dff1161b4 Reduce max_lr_factor from 3.0 to 2.5 Daniel Povey 2022-07-19 05:15:03 +08:00
  • 8bb44b2944 Change param_pow from 0.6 to 0.7 Daniel Povey 2022-07-19 05:08:32 +08:00
  • 749dfcdb7d changes for RESULTS.md luomingshuang 2022-07-18 22:08:42 +08:00
  • 5a6f8fb22a add other decoding methods for wenetspeech luomingshuang 2022-07-18 22:05:53 +08:00
  • d773b29db2 Merge branch 'master' into streaming5 pkufool 2022-07-18 18:09:10 +08:00
  • 608473b4eb
    Add RNN-LM rescoring in fast beam search (#475) ezerhouni 2022-07-18 10:52:17 +02:00
  • 1a5f9f7c74 Replace HLG to LG in docstring Erwan 2022-07-18 10:19:14 +02:00
  • 7bd6be17e8
    Merge 1ebf714fb758942266ef8a8fdcae54c5061f762c into aec222e2fe96bba7b2a7c96bcb2327a2fd45dfdc LIyong.Guo 2022-07-18 10:17:06 +02:00
  • 30dd1479fd Add comments for rnn rescoring Erwan 2022-07-18 09:07:27 +02:00
  • 0738a44a46 Merge branch 'conformer_ctc2' of github.com:WayneWiser/icefall into conformer_ctc2 Quandwang 2022-07-18 14:44:26 +08:00
  • 9b5928d3e3
    Merge branch 'k2-fsa:master' into conformer_ctc2 Quandwang 2022-07-18 14:42:51 +08:00
  • aec222e2fe
    add compile_lg.py for aishell2 recipe (#481) Mingshuang Luo 2022-07-18 14:36:40 +08:00
  • 647e1408b6 Merge branch 'conformer_ctc2' of github.com:WayneWiser/icefall into conformer_ctc2 Quandwang 2022-07-18 14:35:51 +08:00
  • 8d7acf9a79 resolve flake8 conflicts Quandwang 2022-07-18 14:34:17 +08:00
  • 21f2f05dba
    Merge branch 'master' into conformer_ctc2 Quandwang 2022-07-18 14:24:43 +08:00
  • f3618c989a remove unnecessary func Quandwang 2022-07-18 14:20:40 +08:00
  • ae28572fb0 add compile_lg.py for aishell2 recipe luomingshuang 2022-07-18 14:13:22 +08:00