Commit Graph

  • 8e454bcf9e Exclude size=500 dim from projection; try to use double for model average Daniel Povey 2022-05-26 15:15:27 +08:00
  • 364bccb2e3 Merge branch 'master' into streaming-conformer pkufool 2022-05-26 10:29:08 +08:00
  • 4f2ef237eb streaming for pruned_transducer_stateless4 pkufool 2022-05-26 10:22:05 +08:00
  • e923b1b336 Minor fixes pkufool 2022-05-26 10:00:18 +08:00
  • b056e55ead
    Merge 74200583be92b96726de086023db2a0f45a1d401 into 8eff7a4642da89bace5ea60ea9c5cae7ef5043b0 Zengwei Yao 2022-05-25 09:48:59 +00:00
  • 74200583be minor change of the matrix_bd_utterance yaozengwei 2022-05-25 17:12:47 +08:00
  • 435a01dbdf add doc about memory yaozengwei 2022-05-25 17:09:49 +08:00
  • 8eff7a4642
    Add tail padding to Emformer model. (#384) streaming Fangjun Kuang 2022-05-24 23:29:52 +08:00
  • ff8343e378 Add tail padding to Emformer model. Fangjun Kuang 2022-05-24 23:27:41 +08:00
  • c8c8645081
    [Ready to merge] Pruned-transducer-stateless2 recipe for aidatatang_200zh (#375) Mingshuang Luo 2022-05-24 23:07:40 +08:00
  • 70e302cf2b
    First attempt to add WEB client to the streaming emformer. (#351) Fangjun Kuang 2022-05-24 17:16:00 +08:00
  • 8c5722de8c
    [egs] Add prefix when reading manifests due to recent lhotse changes (#382) Ewald Enzinger 2022-05-23 17:37:35 +02:00
  • 0e50e5bde7 Fix style issues Ewald Enzinger 2022-05-23 17:08:58 +02:00
  • fd224a5eca Fix wenetspeech Ewald Enzinger 2022-05-23 15:33:52 +02:00
  • 656ed06390 [egs] Add prefix when reading manifests due to recent lhotse changes Ewald Enzinger 2022-05-23 15:26:57 +02:00
  • 4e94bf7407 sovle conflicts luomingshuang 2022-05-23 20:10:24 +08:00
  • ca61f189be Add teacher student loss. Fangjun Kuang 2022-05-23 19:15:45 +08:00
  • 3e2dbc9ab5 Minor fixes. Fangjun Kuang 2022-05-23 18:18:04 +08:00
  • b7676ca1f2 Save averaged models periodically during training Fangjun Kuang 2022-05-23 18:15:55 +08:00
  • bf3df442c6 Merge remote-tracking branch 'dan/master' into rnnt-lstm-2022-04-21 Fangjun Kuang 2022-05-23 17:33:26 +08:00
  • 0e57b30495
    [Ready to merge] Pruned Transducer Stateless2 for WenetSpeech (char-based) (#349) Mingshuang Luo 2022-05-23 17:13:01 +08:00
  • 0d8fa6ae65 code style check luomingshuang 2022-05-23 16:08:49 +08:00
  • 2f1e23cde1
    Narrower and deeper conformer (#330) Fangjun Kuang 2022-05-23 14:39:11 +08:00
  • 478bc42910 Minor fixes. Fangjun Kuang 2022-05-23 14:38:00 +08:00
  • a896d982ec Update results. Fangjun Kuang 2022-05-23 14:25:58 +08:00
  • d618b005fc Merge remote-tracking branch 'dan/master' into deeper-conformer Fangjun Kuang 2022-05-23 14:12:50 +08:00
  • 935966ba94 Revert to use random combiner. Fangjun Kuang 2022-05-23 14:10:49 +08:00
  • 68a82cbd22 Minor fixes. Fangjun Kuang 2022-05-23 12:53:40 +08:00
  • 7cc697c03a support streaming on pruned_transducer_stateless2; add delay penalty; fixes for decode states pkufool 2022-05-22 17:14:17 +08:00
  • 9ef11e64ba Some small fixes, to bias_correction2 formula and remove bias-u,v-scale Daniel Povey 2022-05-22 16:28:33 +08:00
  • b916789ca3 Further increase scales Daniel Povey 2022-05-22 12:25:26 +08:00
  • 9e206d53fc Increase initial scale for conv and self_attn Daniel Povey 2022-05-22 12:18:57 +08:00
  • 56d9928934 Scale down modules at initialization Daniel Povey 2022-05-22 11:56:59 +08:00
  • 5d57dd3930 Change initial bias scales from 0.1 to 0.2 Daniel Povey 2022-05-22 10:59:51 +08:00
  • 435b073979 Change init of biases to all -0.1..0.1 Daniel Povey 2022-05-22 10:43:06 +08:00
  • fb8e926521 fix decode.py yaozengwei 2022-05-21 20:53:40 +08:00
  • cbdddbe670 Reduce scale on bias from 4 to 2 Daniel Povey 2022-05-21 20:07:41 +08:00
  • dcab1aee4e make biases learn faster in a different way. Daniel Povey 2022-05-21 17:47:30 +08:00
  • 992f922f81 Make biases learn at twice the speed Daniel Povey 2022-05-21 17:30:45 +08:00
  • b7adb6d738 Slow down learning of simple_am_proj.weight by 0.5 Daniel Povey 2022-05-21 17:06:24 +08:00
  • 11eac9089e Fix w.r.t. how cov stats are estimated: storing more stats, and now always zeroing the stats. Daniel Povey 2022-05-20 23:05:05 +08:00
  • 3806d2db19 fix bug in ConvolutionModule yaozengwei 2022-05-20 22:28:41 +08:00
  • ec5a112831
    [Ready to merge] Do some coding style checks for the latest files (#379) Mingshuang Luo 2022-05-20 19:30:38 +08:00
  • 97003f1760 a change for compute_fbank_yesno.py luomingshuang 2022-05-20 18:51:00 +08:00
  • df7142de77 do changes for .flake8 luomingshuang 2022-05-20 18:15:19 +08:00
  • 0b4a8d6c53 style check luomingshuang 2022-05-20 18:10:47 +08:00
  • 6f974b32f6 Restore missing factor 1-beta1 Daniel Povey 2022-05-20 17:43:48 +08:00
  • 768c260a4d Slightly simplify scaling code Daniel Povey 2022-05-20 16:43:05 +08:00
  • 118b09463d Add more documents pkufool 2022-05-20 15:26:45 +08:00
  • 4257c1024f Remove initial_speed option Daniel Povey 2022-05-20 13:41:56 +08:00
  • abe5abb688 Implement Cain with scdaling incorporated; Removing scaling from ScaledLinear, ScaledConv1d, etc. Daniel Povey 2022-05-20 13:36:01 +08:00
  • 19a8700301 add pos enc on memory, left context, chunk, and right context (in key) yaozengwei 2022-05-20 11:13:29 +08:00
  • 5e507f3514 change for .flake8 luomingshuang 2022-05-20 01:47:19 +08:00
  • fa83775d98 add pruned-rnnt2 recipe for alimeeting dataset luomingshuang 2022-05-20 01:42:10 +08:00
  • afdda0509c solve the conflicts luomingshuang 2022-05-20 00:52:12 +08:00
  • 3925eaeb15 a small change for .flake8 luomingshuang 2022-05-20 00:48:27 +08:00
  • 465803e219 Minor fixes. Fangjun Kuang 2022-05-20 00:43:16 +08:00
  • 34f859fd92 solve the conflicts luomingshuang 2022-05-20 00:42:55 +08:00
  • bcd0e872b8 Add more comments Fangjun Kuang 2022-05-20 00:38:44 +08:00
  • 7fb31c1eea do some changes luomingshuang 2022-05-20 00:32:27 +08:00
  • 4b59573473 Add non-streaming ASR server implementation Fangjun Kuang 2022-05-20 00:21:30 +08:00
  • 8fd9e64fdf Fix eps value Daniel Povey 2022-05-19 22:25:18 +08:00
  • 1edc0fa841 Fixes to cain Daniel Povey 2022-05-19 22:21:41 +08:00
  • 6085ab64ef Make cain average over more iters and use preconditioning on the other dims first Daniel Povey 2022-05-19 21:34:12 +08:00
  • 6f2f7aff73 do some changes luomingshuang 2022-05-19 14:57:42 +08:00
  • 884bf615e5 change for README.md luomingshuang 2022-05-19 14:25:24 +08:00
  • 6c8c8b5faa do some changes luomingshuang 2022-05-19 14:22:05 +08:00
  • 2900ed8f8f
    Merge pull request #376 from danpovey/diagnostics_fix Daniel Povey 2022-05-19 12:51:07 +08:00
  • ac5a9faafd Merge branch 'master' into knowledge_base_1b_L2_ng_orth2feat_cain Daniel Povey 2022-05-19 12:49:24 +08:00
  • 9e88d0bf31 Merge remote-tracking branch 'upstream/master' Daniel Povey 2022-05-19 12:49:12 +08:00
  • 5230e73e41 Small fixes Daniel Povey 2022-05-19 12:49:00 +08:00
  • e44edf99a4 Merge branch 'master' into knowledge_base_1b_L2_ng_orth2feat_cain Daniel Povey 2022-05-19 12:42:50 +08:00
  • b33b2e5192 add pruned-rnnt2 model for aidatatang_200zh luomingshuang 2022-05-19 12:33:05 +08:00
  • 4e23fb2252
    Improve diagnostics code memory-wise and accumulate more stats. (#373) Daniel Povey 2022-05-19 11:45:59 +08:00
  • c736b39c7d Remove unnecessary option for diagnostics code, collect on more batches Daniel Povey 2022-05-19 11:35:54 +08:00
  • c0fdfabaf3 Remove memory-limit options arg Daniel Povey 2022-05-19 11:30:56 +08:00
  • c2c46ea023 Update diagnostics, hopefully print more stats. Daniel Povey 2022-05-18 21:42:57 +08:00
  • 5bd2490b44 support streaming in conformer pkufool 2022-05-18 23:26:24 +08:00
  • ebc2ffeff7 Bug fix Daniel Povey 2022-05-18 10:26:29 +08:00
  • 668d01cc7a Replace Eve with Cain in pruned_transducer_stateless4 Daniel Povey 2022-05-18 10:16:50 +08:00
  • 38ccbb6062 Think it works now.. will move to different branch though. Daniel Povey 2022-05-18 10:11:20 +08:00
  • 7b2239352c add pretrained to pruned_transducer_stateless2 Gaurav Shegokar 2022-05-17 22:39:26 +00:00
  • 170768082f
    Merge 6d809bad0bfe6ea1beb36eabf7c71e2c5f7716e9 into f6ce135608968bfe736040345743dfbc2847cd8b Fangjun Kuang 2022-05-17 13:00:43 -04:00
  • 3d35018ce0 Just adding some draft code, will move this around. Daniel Povey 2022-05-17 21:00:36 +08:00
  • a46f74feb4 Broadcast transformation Daniel Povey 2022-05-17 17:22:43 +08:00
  • ceb4eb4b85 Scale covariance Daniel Povey 2022-05-17 15:32:06 +08:00
  • c923b5900e Symmetrize covariance Daniel Povey 2022-05-17 15:14:23 +08:00
  • 9133b57808 Use torch.no_grad() for stats Daniel Povey 2022-05-17 14:49:00 +08:00
  • 07d3369234 Finish feat-diagonalizing code Daniel Povey 2022-05-17 14:13:56 +08:00
  • e1b9211de2 update results based on dynamicbucketsampler luomingshuang 2022-05-17 11:58:03 +08:00
  • 822fe21a83 use dynamicbucketsampler for decoding luomingshuang 2022-05-17 09:38:25 +08:00
  • cc678861d4 Merge branch 'master' of https://github.com/k2-fsa/icefall Gaurav Shegokar 2022-05-16 17:05:08 +00:00
  • ef56cd87e4 pretrain scriptping Gaurav Shegokar 2022-05-16 17:04:57 +00:00
  • 8301faeb12 Fix an error. Fangjun Kuang 2022-05-16 22:31:36 +08:00
  • 74b3861320 add RESULTS.md luomingshuang 2022-05-16 22:19:38 +08:00
  • f6ce135608
    Various fixes to support torch script. (#371) Fangjun Kuang 2022-05-16 21:46:59 +08:00
  • 3d833d9430 Add epsilon self-loops to G. Fangjun Kuang 2022-05-16 21:45:20 +08:00
  • 65bca6ea90 change for README.md luomingshuang 2022-05-16 21:18:47 +08:00
  • 42b8ce9244 change for .flake8 luomingshuang 2022-05-16 21:07:58 +08:00
  • cd3bf0a1fd do some changes luomingshuang 2022-05-16 20:53:19 +08:00