“Sinovoice-2016-5-5”版本间的差异

来自cslt Wiki
跳转至： 导航、 搜索
（→‎Big-Model Training）
 
第107行： 第107行：
 
* making lattice for MPE training.
 
* making lattice for MPE training.
  
−
===SinSong Robot===
+
===Character LM===
−
* Test based on 10000h(7*2048-xent) model
+
* Except Sogou-2T, 9-gram has been done.
−
  ------------------------------------------------
+
* Add word boundary tag to Character-LM trainig done
−
    condition | clean  | replay(0.5m) | real-env
+
:* 9-gram
−
  ------------------------------------------------
+
:* Except Weibo & Sogou-2T
−
      wer    |  3    |  18(mpe-14)  | too-bad
+
* Prepare specific domain vocabulary
−
  ------------------------------------------------
+
:* Dianxin/Baoxian/Dianli
  
−
* Plan to record in restaurant on April 10.
+
*DT lm training
−
 
+
−
===Character LM===
+
−
*Except Sogou-2T, 9-gram has been done.
+
−
*Worse than word-lm(9%->6%)
+
−
*Add word boundary tag to Character-LM trainig
+
 
*Merge Character-LM  & word-LM
 
*Merge Character-LM  & word-LM
 
:* Union
 
:* Union
 
:* Compose, success.
 
:* Compose, success.
 
* 2-step decoding: first, character-based LM. Then, word-based LM.
 
* 2-step decoding: first, character-based LM. Then, word-based LM.
−
*Word boundary character training
+
 
 +
===SiaSong Robot===
 +
* Beam-forming algorithm test
 +
* NN-model based beam-forming
  
 
===Project===
 
===Project===
第134行： 第132行：
 
==SID==
 
==SID==
 
===Digit===
 
===Digit===
−
* DNN-PLDA gets better performance than i-Vector;
+
* Engine Package
−
DNN
+
−
cosine
+
−
10.4167%, at threshold 89.3973
+
−
9.72222%, at threshold 87.8146
+
−
8.68056%, at threshold 84.2021
+
−
3.47222%, at threshold 11.5852
+
−
lda
+
−
3.125%, at threshold 54.1172
+
−
2.77778%, at threshold 50.1447
+
−
2.43056%, at threshold 48.6887
+
−
1.73611%, at threshold 14.5075
+
−
plda
+
−
2.43056%, at threshold -23.954
+
−
2.08333%, at threshold -24.6051
+
−
2.08333%, at threshold -21.0524
+
−
1.73611%, at threshold 4.83949
+
−
 
+
−
ivector
+
−
plda
+
−
3.15789%, at threshold 0.563044
+
−
3.85965%, at threshold 0.525273
+
−
3.85965%, at threshold 0.502531
+
−
2.80702%, at threshold 0.429186
+

2016年5月5日 (四) 06:35的最后版本

Data

  • 16K LingYun
  • 2000h data ready
  • 4300h real-env data to label
  • YueYu
  • Total 250h(190h-YueYu + 60h-English)
  • Add 60h YueYu
  • CER: 75%->76%
  • WeiYu
  • 50h for training
  • 120h labeled ready

Model training

Deletion Error Promblem

  • Add one noise phone to alleviate the silence over-training
  • Omit sil accuracy in discriminative training
  • H smoothing of XEnt and MPE
  • Testdata: test_1000ju from 8000ju 
  -----------------------------------------------------------------------------
                 model                    | ins  |  del  | sub | wer/tot-err  
  -----------------------------------------------------------------------------
   svd600_lr2e-5_1000H_mpe_uv-fix         |  24  |  56   | 408 | 8.26/488
  -----------------------------------------------------------------------------
svd600_lr2e-5_1000H_mpe_uv-fix_omitsilacc |  32  |  48   | 409 | 8.28/489
  -----------------------------------------------------------------------------
   svd600_lr2e-5_1000H_mpe_uv-fix_xent0.1 |  24  |  57   | 406 | 8.24/487
  -----------------------------------------------------------------------------
   svd600_lr2e-5_1000H_mpe_uv-fix_xent0.2 |  25  |  60   | 409 | 8.36/494
  -----------------------------------------------------------------------------
  • Testdata: test_8000ju 
  -----------------------------------------------------------------------------
                 model                    | ins  |  del  | sub  | wer/tot-err  
  -----------------------------------------------------------------------------
   svd600_lr2e-5_1000H_mpe_uv-fix         |  140 |  562  | 3686 | 9.19/4388     | 47753-total-word
  -----------------------------------------------------------------------------
   svd600_lr2e-5_1000H_mpe_uv-fix_xent0.1 |  146 |  510  | 3705 | 9.13/4361
  -----------------------------------------------------------------------------
   svd600_lr2e-5_1000H_mpe_uv-fix_xent0.2 |  139 |  492  | 3739 | 9.15/4370
  -----------------------------------------------------------------------------
  • Testdata: test_2000ju from 10000ju 
  -----------------------------------------------------------------------------
                 model                    | ins  |  del  |  sub | wer/tot-err  
  -----------------------------------------------------------------------------
   svd600_lr2e-5_1000H_mpe_uv-fix         |  86  |  790  | 1471 | 18.55/2347
  -----------------------------------------------------------------------------
svd600_lr2e-5_1000H_mpe_uv-fix_omitsilacc |  256 |  473  | 1669 | 18.95/2398
  -----------------------------------------------------------------------------
   svd600_lr2e-5_1000H_mpe_uv-fix_xent0.1 |  95  |  704  | 1548 | 18.55/2347
  -----------------------------------------------------------------------------
   svd600_lr2e-5_1000H_mpe_uv-fix_xent0.2 |  100 |  697  | 1557 | 18.60/2354
  -----------------------------------------------------------------------------
  • Testdata: test_10000ju 
  -----------------------------------------------------------------------------
                 model                    | ins  |  del  | sub  | wer/tot-err  
  -----------------------------------------------------------------------------
   svd600_lr2e-5_1000H_mpe_uv-fix         |  478 | 3905  | 7698 | 18.31/12081  | 65989-total-word
  -----------------------------------------------------------------------------
   svd600_lr2e-5_1000H_mpe_uv-fix_xent0.1 |  481 | 3741  | 7773 | 18.18/11995
  -----------------------------------------------------------------------------
   svd600_lr2e-5_1000H_mpe_uv-fix_xent0.2 |  502 | 3657  | 7826 | 18.16/11985
  -----------------------------------------------------------------------------
  • Add one silence arc from start-state to end-state

Big-Model Training

  • 16k
  • 8k 
PingAnAll:
 ==================================================================================
 |         AM / error         | tot_err |   ins   |   del   |   sub   |   wer   |
 ----------------------------------------------------------------------------------
 | tdnn 7-1024 xEnt 2500.mdl  |  3626   |   619   |   773   |   2234  |  16.60  |
 ----------------------------------------------------------------------------------
 | spn 7-1024 xEnt 300.mdl    |  3746   |   702   |   763   |   2281  |  17.15  |
 ==================================================================================

PingAnUser:
 ==================================================================================
 |         AM / error         | tot_err |   ins   |   del   |   sub   |   wer   |
 ----------------------------------------------------------------------------------
 | tdnn 7-1024 xEnt 2500.mdl  |   549   |   158   |    75   |   316   |  35.91  |
 ----------------------------------------------------------------------------------
 | spn 7-1024 xEnt 300.mdl    |   571   |   151   |    97   |   323   |  37.34  |
 ==================================================================================

LiaoNingYiDong:
 ==================================================================================
 |         AM / error         | tot_err |   ins   |   del   |   sub   |   wer   |
 ----------------------------------------------------------------------------------
 | tdnn 7-1024 xEnt 2500.mdl  |   5873  |   879   |   1364  |   3630  |  21.72  |
 ----------------------------------------------------------------------------------
 | spn 7-1024 xEnt 300.mdl    |   6257  |   977   |   1348  |   3923  |  23.14  |
 ==================================================================================

Embedding

  • The size of nnet1 AM is 6.4M (3M after decomposition). So we need to control AM size within 10M.
  • 5*576-2400 TDNN model training done. AM size is about 17M
  • 5*500-2400 TDNN model on training.
  • making lattice for MPE training.

Character LM

  • Except Sogou-2T, 9-gram has been done.
  • Add word boundary tag to Character-LM trainig done
  • 9-gram
  • Except Weibo & Sogou-2T
  • Prepare specific domain vocabulary
  • Dianxin/Baoxian/Dianli
  • DT lm training
  • Merge Character-LM & word-LM
  • Union
  • Compose, success.
  • 2-step decoding: first, character-based LM. Then, word-based LM.

SiaSong Robot

  • Beam-forming algorithm test
  • NN-model based beam-forming

Project

  • Pingan & Yueyu Deletion error too more
  • TDNN deletion error rate > DNN deletion error rate
  • TDNN Silence scale is too sensitive for different test cases.

SID

Digit

  • Engine Package
取自“http://index.cslt.org/mediawiki/index.php?title=Sinovoice-2016-5-5&oldid=20083”