Commit Graph

2712 Commits

Author SHA1 Message Date
游雁
7699a35d2c v1.1.12 2024-10-29 15:13:14 +08:00
Truco
1a45b647a8
perf(models/FsmnVADStreaming): optimize GetFrameState and PopDataToOutputBuf (#2177)
- In GetFrameState(), pass generator to sum() instead of generating a list, ~10% gain in a 21s sample
- In GetFrameState(), cast `sum_score` (a tensor) to float to reduce calling to tensor lib,
  ~13% gain in a 23s example
- In PopDataToOutputBuf(), remove unused `out_pos` and related calculation, ~10% gain in a 27s sample
2024-10-28 13:41:38 +08:00
StevenH
1254e8aee1
optimize ComputeDecibel in fsmn-vad model by using numpy (#2174) 2024-10-26 12:19:07 +08:00
Vignesh Skanda
e6fe606577
Update register.py (#2145) 2024-10-16 13:45:22 +08:00
Kun Lu
db308e7535
feat: add campplus merge_thr (#2135) 2024-10-15 17:52:10 +08:00
pointerhacker
70645e4807
数据并行可能导致的模型训练报错 (#2139)
* fix: 修复数据并行训练中ä¼可能会出现的错误

* fix: 修复数据并行训练中ä¼可能会出现的错误

* fix: 修复数据并行ènot need tensor

---------

Co-authored-by: zhaochaojin <zhaochaojin@didiglobal.com>
2024-10-15 17:50:51 +08:00
游雁
5c28b4d612 whisper-large-v3-turbo 2024-10-14 00:21:24 +08:00
游雁
cd68458099 whisper-large-v3-turbo 2024-10-11 16:10:04 +08:00
游雁
6d932da239 whisper-large-v3-turbo 2024-10-11 14:37:27 +08:00
游雁
2330e58f5f bugfix v1.1.11 2024-10-10 15:32:08 +08:00
游雁
5fc1d918aa v1.1.10 2024-10-09 10:58:52 +08:00
Nixon
5dbe6898ca
fix list index out of range error (#2122)
Co-authored-by: nixonjin <nixonjin@tencent.com>
2024-10-09 10:40:58 +08:00
游雁
0bf8edca37 find_unused_parameters 2024-09-30 15:43:13 +08:00
游雁
5d35b3c70b find_unused_parameters 2024-09-30 12:28:16 +08:00
sugarcase
a8f0aad81d
fsmn_kws_mt finetune and inference adapt to right modelscope hub (#2113)
Co-authored-by: pengteng.spt <pengteng.spt@alibaba-inc.com>
2024-09-27 14:16:28 +08:00
lyblsgo
67239ea39b update type_utils 2024-09-25 23:38:22 +08:00
游雁
1a44f82ead v1.1.8 2024-09-25 16:33:37 +08:00
游雁
c0e7b17f08 sensevoice bugfix 2024-09-25 16:32:58 +08:00
游雁
8b0f876c0c v1.1.7 2024-09-25 15:27:06 +08:00
游雁
fc547e14e8 bugfix memory leaky 2024-09-25 15:26:14 +08:00
zhifu gao
2196844d1d
Dev kws (#2105)
* multi tokenizer

* support fsmn_kws, fsmn_kws_mt, sanm_kws, sanm_kws_streaming training

* kws

---------

Co-authored-by: pengteng.spt <pengteng.spt@alibaba-inc.com>
2024-09-25 15:10:50 +08:00
Nixon
1af68ba6ff
fix bug, 1 fix cuda oom, 2 fix choose a window size 400 that is [2, 0] (#2075)
Co-authored-by: nixonjin <nixonjin@tencent.com>
2024-09-14 10:13:23 +08:00
游雁
1b3c0da937 v1.1.6 2024-08-20 13:59:24 +08:00
lingji-yidong
f43da18b5e
fix start timestamp in sentence_info (#2024)
Previously, the start timestamp was defined as the last character's timestamp[0] of the previous sentence. It has now been changed to the first character's timestamp[0] of the current sentence.
2024-08-19 13:36:59 +08:00
游雁
5c11fd761e version check 2024-08-12 11:04:13 +08:00
游雁
a28de72b17 bugfix for paraformer-streaming 2024-08-02 01:55:30 +08:00
游雁
7f6099140a whisper bugfix 2024-08-02 01:20:34 +08:00
游雁
d238a5ab44 dcos 2024-07-30 17:45:13 +08:00
维石
09f0f20545 bug fix 2024-07-29 14:20:01 +08:00
Ziyao Wang
9f04fd130c
Fix invalid escape sequence '\w' in RegEx due to 'r' missing (#1967) 2024-07-29 09:47:45 +08:00
游雁
8b0f765b47 v1.1.4 2024-07-27 00:44:55 +08:00
游雁
69370ed3d6 cache bugfix 2024-07-27 00:44:23 +08:00
北念
8aa01036b6 fix sensevoice loss_rich 2024-07-24 19:09:02 +08:00
北念
f41944daa8 add sensevoice scp2jsonl 2024-07-24 10:00:17 +08:00
北念
1b97842604 add sensevoice scp2jsonl 2024-07-23 11:45:43 +08:00
游雁
37fc6ad946 v1.1.3 2024-07-22 17:04:29 +08:00
维石
2ae59b6ce0 ONNX and torchscript export for sensevoice 2024-07-22 16:58:27 +08:00
gaochangfeng
340c55838b
EMO_UNK禁用和Merge VAD修复 (#1940)
* 添加富文本解码约束

* special token

* bug fix

* fix

* 增加unk score的参数

* emobaned

* kwargs2cfg

* merge_vad bug fix

---------

Co-authored-by: 常材 <gaochangfeng.gcf@alibaba-inc.com>
2024-07-22 15:28:27 +08:00
游雁
f9c13d7f4b bugfix 2024-07-22 15:07:59 +08:00
游雁
e42f539eb8 bugfix 2024-07-22 15:03:35 +08:00
Shi Xian
e09d87193b
Merge pull request #1928 from liugz18/main
Rename 'res' in line 514 to avoid with naming conflict with line 365
2024-07-22 11:32:39 +08:00
凪咲
85c1675e7c
fix: fix input download logic (#1929) 2024-07-19 10:26:58 +08:00
liugz18
d80ac2fd2d
Rename 'res' in line 514 to avoid with naming conflict with line 365 2024-07-18 21:34:55 +08:00
北念
bd352983c6 add default emo and event target for sensevoice 2024-07-18 15:05:28 +08:00
北念
a98550fdf5 fix sense_voice_datasets 2024-07-17 16:05:58 +08:00
游雁
ee8b6e2d99 v1.1.2 2024-07-16 14:22:30 +08:00
游雁
8b74979bd3 sensevoice 2024-07-16 13:57:51 +08:00
游雁
f097706c40 v1.1.1 2024-07-16 10:56:14 +08:00
彭震东
f2ed4b3856
fix progress bar for batch_size (#1917) 2024-07-15 17:54:27 +08:00
北念
5448e926a2 add postprocess for sensevoice 2024-07-10 11:27:35 +08:00