Commit Graph

  • e52581e69b support local_rank for multi-node root 2025-05-16 00:02:12 -07:00
  • 0e8c1db4d0 fix speed perturb issue root 2025-05-15 22:45:04 -07:00
  • bfb4ebeb83 remove triton root 2025-05-15 14:32:49 +00:00
  • f81363d324 add speech continuation pretraining root 2025-05-15 14:16:51 +00:00
  • e6615df4eb fix stage 5 output pathing Kinan Martin 2025-05-15 09:11:40 +09:00
  • daff070d68 restore version of mls_english compute_fbank_mls_english.py and prepare.sh from commit 547f5c5 Kinan Martin 2025-05-15 07:24:26 +09:00
  • e34f2dbb2a merge change to remove bilingual param with new multidataset_datamodule Kinan Martin 2025-05-14 08:51:11 +09:00
  • eb5004880f deprecate params.bilingual=0, replace ReazonSpeechAsrDataModule for MultiDatasetAsrDataModule, not tested yet Kinan Martin 2025-05-14 08:40:15 +09:00
  • 7ef1811063 remove bilingual tag from train.py Bailey Hirota 2025-05-14 08:37:44 +09:00
  • e65725810c fix mmsu root 2025-05-13 09:13:12 +00:00
  • 24b6f42340 fix typos in docs Yifan Yang 2025-05-13 14:34:05 +08:00
  • 62dfe56cbe restore checkpoint save after validation yifanyeung 2025-05-13 06:14:59 +00:00
  • cbf3af31fd add voicebench eval root 2025-05-13 05:37:11 +00:00
  • b2df5bbb83 Revert "add fbank" Bailey Hirota 2025-05-02 23:18:53 +09:00
  • 82bd37cacd add fbank Bailey Hirota 2025-05-02 03:31:55 +09:00
  • 06667e1f6d add batch shave mechanism yfyeung 2025-05-12 16:49:42 +00:00
  • ea20ac208d
    Merge branch 'k2-fsa:master' into dev/speechllm Yifan Yang 2025-05-12 20:31:41 +08:00
  • 407828f22e Support multiple parallel augmentations. Fangjun Kuang 2025-05-12 19:35:37 +08:00
  • e79833aad2
    ensure SwooshL/SwooshR output dtype matches input dtype (#1940) Yifan Yang 2025-05-12 19:28:48 +08:00
  • 89781b9bb1 add cosyvoice2 decode root 2025-05-12 10:06:59 +00:00
  • c709ce433d
    Merge branch 'k2-fsa:master' into dev/speechllm Yifan Yang 2025-05-12 14:38:13 +08:00
  • 2793ccdf56 remove checkpoint save after validation yfyeung 2025-05-12 06:36:20 +00:00
  • 9a0b5d706b
    Merge branch 'k2-fsa:master' into fix/swoosh Yifan Yang 2025-05-12 14:23:19 +08:00
  • 4627969ccd
    fix bug: undefined name 'partial' (#1941) Yifan Yang 2025-05-12 14:19:53 +08:00
  • d3f9603e58
    fix bug: undefined name 'partial' Yifan Yang 2025-05-12 13:16:49 +08:00
  • e99e621f65
    ensure SwooshL/SwooshR output dtype matches input dtype Yifan Yang 2025-05-12 12:58:57 +08:00
  • c078772e59 skip OOM yfyeung 2025-05-11 17:23:19 +00:00
  • 9939c2b72d remove duplicated torch autocast yfyeung 2025-05-11 17:03:44 +00:00
  • 5fbeed9f96
    fix SwooshR and SwooshL Yifan Yang 2025-05-12 00:48:42 +08:00
  • cd3adad46d use quadratic-duration yfyeung 2025-05-10 17:47:05 +00:00
  • c75767f600 set world_size and rank explicitly yfyeung 2025-05-10 17:32:14 +00:00
  • fd31ed5b0b
    minor fix ssl_datamodule.py Yifan Yang 2025-05-11 00:59:06 +08:00
  • 260d37b65a
    Merge branch 'k2-fsa:master' into dev/k2ssl Yifan Yang 2025-05-11 00:33:35 +08:00
  • 2420d0c95f
    update multi_dataset.py Yifan Yang 2025-05-10 02:13:25 +08:00
  • ec6c8f748d fix data prepare yfyeung 2025-05-09 17:18:22 +00:00
  • b20a0d0e35 add on the fly feature root 2025-05-08 19:21:41 -07:00
  • 21d1bf73bb new version of multi_ja_en prepare.sh script which swaps Librispeech for MLS English Kinan Martin 2025-05-09 10:57:41 +09:00
  • 489c42b45e support zipformer encoder Yifan Yang 2025-05-08 04:31:34 +00:00
  • bd2df570ad add debug script root 2025-05-08 03:37:26 -07:00
  • 37db65984c remove k2 dependency root 2025-05-08 03:02:34 -07:00
  • e41c1cabd5 add dependency root 2025-05-08 07:56:14 +00:00
  • 7cc366d82d add en data, cosy2 token for training root 2025-05-08 07:23:22 +00:00
  • 2dd40b62ef add vocalnet en data root 2025-05-08 06:29:46 +00:00
  • 211c01bc1d format train.py Yifan Yang 2025-05-07 12:37:19 +00:00
  • 23b5a7ce3e format multi_dataset.py Yifan Yang 2025-05-07 12:29:12 +00:00
  • 9c8c4314de init zipformer_llm_zh Yifan Yang 2025-05-07 12:18:41 +00:00
  • 547f5c5cfb optimize with num_jobs on save_audios Kinan Martin 2025-05-02 07:22:38 +09:00
  • 88249f0eb4 fix stage 2 and 3 Kinan Martin 2025-05-01 08:15:07 +09:00
  • 90326c1f43 fix validation manifest name Kinan Martin 2025-05-01 08:05:42 +09:00
  • dc07bba236 init Yifan Yang 2025-04-30 09:54:42 +00:00
  • cd7caf12df
    Fix speech_llm recipe (#1936) Yifan Yang 2025-04-30 11:41:00 +08:00
  • b21f3aace5 Merge branch 'fix/speechllm' of github.com:yfyeung/icefall into fix/speechllm Your Name 2025-04-29 20:34:22 -07:00
  • 1210025d8b fix flake8 Your Name 2025-04-29 20:34:12 -07:00
  • dbe270ba94 adjusted prepare.sh to only calculate fbank and manifest together; adjust datamodule to load from manifest files Kinan Martin 2025-04-30 10:06:13 +09:00
  • bef49cca86
    Merge branch 'k2-fsa:master' into fix/speechllm Yifan Yang 2025-04-30 01:44:58 +08:00
  • 973fcbc585 add missing backslash yfyeung 2025-04-29 10:39:29 -07:00
  • 26c022665b remove space if existing yfyeung 2025-04-29 10:32:12 -07:00
  • d1c336f589 remove ineffective normalize_text_alimeeting yfyeung 2025-04-29 10:28:01 -07:00
  • 26aef4a926 remove ineffective normalize_text_alimeeting yfyeung 2025-04-29 06:24:43 -07:00
  • 46b9be31cc remove batch shaving yfyeung 2025-04-28 04:37:57 -07:00
  • f5d2aa1f5d fix train/eval mode yfyeung 2025-04-28 09:10:12 +00:00
  • 59c577f4ef Fix convert_texts_into_ids() in the tedlium3 recipe. (#1929) Fangjun Kuang 2025-04-24 17:04:46 +08:00
  • 08be51a91f change pic root 2025-04-29 10:09:57 +00:00
  • 11bd3c9ad8 lint root 2025-04-29 09:46:44 +00:00
  • 360f0aa397 update README root 2025-04-29 08:49:12 +00:00
  • 448a4eeea7 update hf dataset loading into lhotse root 2025-04-29 07:33:34 +00:00
  • d742043e75 refactor decode part Yuekai Zhang 2025-04-25 18:31:43 +08:00
  • 71a0a442a6 add history cache root 2025-04-25 10:05:07 +00:00
  • 47920c2336 add gradio demo Yuekai Zhang 2025-04-25 16:05:37 +08:00
  • 72addd40f5 change place Yuekai Zhang 2025-04-25 14:22:16 +08:00
  • 9a07363a8d remove unsed Yuekai Zhang 2025-04-25 14:21:50 +08:00
  • 6ea7ec8543 remove offline tab Yuekai Zhang 2025-04-25 14:10:30 +08:00
  • 6955639d22 add qwen omni web demo Yuekai Zhang 2025-04-25 14:08:25 +08:00
  • 3642dfd8c3 refactor code root 2025-04-25 05:36:18 +00:00
  • cc2e64a6aa
    Fix convert_texts_into_ids() in the tedlium3 recipe. (#1929) Fangjun Kuang 2025-04-24 17:04:46 +08:00
  • 2e9be46703 debug root 2025-04-24 08:24:11 +00:00
  • d1312e4d3e Fix convert_texts_into_ids() in the tedlium3 recipe. Fangjun Kuang 2025-04-24 14:50:51 +08:00
  • cf425173af move compute_fbank_mls_english.py, add validate_manifest.py, add shared symlink to librispeech Kinan Martin 2025-04-24 09:39:54 +09:00
  • 3511b7db12 fix train.py and decode.py Your Name 2025-04-23 02:32:05 -07:00
  • 5ec95e5482
    Fix SpeechLLM recipe (#1926) Yifan Yang 2025-04-23 16:18:38 +08:00
  • 64722e7073 fix typos Your Name 2025-04-23 01:17:32 -07:00
  • 58e0016925 fix prepare.sh Your Name 2025-04-23 00:52:04 -07:00
  • 478d56efd8 fix bugs when padding right root 2025-04-23 07:33:27 +00:00
  • c4432d1e35 chmod +x prepare.sh Your Name 2025-04-23 00:25:14 -07:00
  • 4f743993ef instead of on-the-fly features, precompute fbank and manifests in prepare.sh Kinan Martin 2025-04-23 10:13:15 +09:00
  • 23fdef2fd3 add codec decode Yuekai Zhang 2025-04-21 17:57:57 +08:00
  • 09d81b44a7 change padding side name Yuekai Zhang 2025-04-21 17:10:25 +08:00
  • 7db40052d6 add flash attn support Yuekai Zhang 2025-04-21 14:54:28 +08:00
  • b305cdacc0 fix padding side root 2025-04-21 06:23:10 +00:00
  • bdb60f6ddc add codec lm root 2025-04-21 01:00:06 +00:00
  • c49c3f2634 remove musan-k2-v2-reazonspeech-medium from repo Bailey Hirota 2025-04-16 13:30:32 +09:00
  • cb4f4a941b complete musan rezonspeech integration Bailey Hirota 2025-04-16 13:25:40 +09:00
  • 4e2a4fdcd8 readme Kinan Martin 2025-04-16 08:13:59 +09:00
  • bb6d672b54 pre-commit hooks Kinan Martin 2025-04-16 08:05:05 +09:00
  • e69e1c04b2 separate transcript prep stage from bpe train stage Kinan Martin 2025-04-16 07:15:25 +09:00
  • 6e81d9aa5b symlink copied files to librispeech recipe dir Kinan Martin 2025-04-16 07:10:39 +09:00
  • 0e868049a6
    Merge branch 'k2-fsa:master' into mls_english_clean Kinan Martin 2025-04-15 17:52:18 -04:00
  • 458d697acc fix batch_size>1 decoding bug root 2025-04-15 13:41:33 +00:00
  • 0c02da82ac refine decoding method root 2025-04-15 06:53:20 +00:00
  • 3ad075af60 s2t training root 2025-04-15 02:16:03 +00:00