“Sinovoice-2016-5-26”版本间的差异

来自cslt Wiki
跳转至： 导航、 搜索
（→‎Project）
Zxw（讨论 | 贡献）
 
（1位用户的9个中间修订版本未显示）
第18行： 第18行：
 
:*100h User data done
 
:*100h User data done
  
−
==Model training==
+
==Model Training==
−
==Deletion Error Promblem==
+
===Deletion Error Problem===
−
* Add one noise phone to alleviate the silence over-training
+
* Add one noise phone to alleviate the silence over-training, looks OK.
 
* Omit sil accuracy in discriminative training
 
* Omit sil accuracy in discriminative training
−
* H smoothing of XEnt and MPE
+
* H smoothing of XEnt and MPE, no significant affect.
 
* Add one silence arc from start-state to end-state
 
* Add one silence arc from start-state to end-state
  
 
===Big-Model Training===
 
===Big-Model Training===
 
====16k====
 
====16k====
−
 
+
* Done!
 
====8k=====
 
====8k=====
−
=====Model=====
 
−
* Add noise phone
 
−
:* 1300.mdl done
 
−
 
−
* CNN + TDNN
 
−
:* 280/900 mdl
 
−
:* Need about 12 days Xent
 
  
 
=====Project=====
 
=====Project=====
 
* PingAn
 
* PingAn
−
:* Add noise phone to phone-list
 
  
 
   =========================================================================================
 
   =========================================================================================
第62行： 第54行：
 
   | spn 7-1024 MPE adapt-PA_user Hs 2e-5|    15.32    |    32.24    ||    20.93    |
 
   | spn 7-1024 MPE adapt-PA_user Hs 2e-5|    15.32    |    32.24    ||    20.93    |
 
   =========================================================================================
 
   =========================================================================================
 +
 +
  =====================================================================================================
 +
  |    LM / config                          |    KeHu      |KeHu check_zxm_recheck|  KeHu final  |
 +
  -----------------------------------------------------------------------------------------------------
 +
  |    baseline                              |    20.14    |        19.40        |    18.26    |
 +
  -----------------------------------------------------------------------------------------------------
 +
  |bank+baoxian.chart+word.w0.9              |      -      |        19.27        |    17.88    |
 +
  |bank+baoxian+guojiadianwang.chart+word.w0.9|      -      |        19.02        |    17.94    |
 +
  |bank+baoxian+guojiadianwang_w0.9          |      -      |        19.02        |      -      |
 +
  |bank+baoxian_w0.9                          |      -      |        19.20        |      -      |
 +
  |baoxian+bank_w0.9                          |      -      |        19.02        |      -      |
 +
  |baoxian+user200h.chart+word.w0.9          |      -      |        19.20        |      -      |
 +
  |baoxian+user200h_w0.8                      |      -      |        19.02        |      -      |
 +
  |baoxian+user200h_w0.9                      |      -      |        19.08        |    18.01    |
 +
  |baoxian+user200h.w0.9w0.9                  |      -      |        19.08        |      -      |
 +
  |baoxian+user200h.chart1e-7.w0.9w0.1        |      -      |          -          |    23.82    |
 +
  |baoxian+user200h.chart1e-7.w0.9w0.9        |      -      |          -          |    18.26    |
 +
  =====================================================================================================
  
  
第118行： 第128行：
 
* 2-step decoding: first, character-based LM. Then, word-based LM.
 
* 2-step decoding: first, character-based LM. Then, word-based LM.
  
−
===SiaSong Robot===
+
===Problem===
−
* Beam-forming algorithm test
+
* Pingan & Yueyu too much deletion error.
−
* NN-model based beam-forming
+
−
 
+
−
===Project===
+
−
* Pingan & Yueyu Deletion error too more
+
 
:* TDNN deletion error rate > DNN deletion error rate
 
:* TDNN deletion error rate > DNN deletion error rate
 
:* TDNN Silence scale is too sensitive for different test cases.
 
:* TDNN Silence scale is too sensitive for different test cases.
 +
* cmvn causes performance reduction.
 +
 +
==SiaSun Robot==
 +
* Beam-forming algorithm test
 +
* NN-model based beam-forming
  
 
==SID==
 
==SID==
 
===Digit===
 
===Digit===
 
* Engine Package
 
* Engine Package

2016年5月26日 (四) 06:30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
  • 8k more data
  • 50h for training
  • 120h labeled ready
  • PingAn
  • 100h User data done

Model Training

Deletion Error Problem

  • Add one noise phone to alleviate the silence over-training, looks OK.
  • Omit sil accuracy in discriminative training
  • H smoothing of XEnt and MPE, no significant affect.
  • Add one silence arc from start-state to end-state

Big-Model Training

16k

  • Done!

8k=

Project
  • PingAn 
 =========================================================================================
 |     AM / config                     |      all      |    KeHu wer   ||  KeHu no-ins  |
 -----------------------------------------------------------------------------------------
 | tdnn 7-2048 xEnt                    |     16.45     |     36.49     ||     25.18     |
 | tdnn 7-2048 MPE                     |     15.22     |     32.77     ||     23.48     |
 | tdnn 7-2048 MPE adapt-PABX          |     14.67     |     31.33     ||     22.76     |
 -----------------------------------------------------------------------------------------
 | tdnn 7-1024 xEnt                    |     16.60     |     35.91     ||     25.58     |
 | tdnn 7-1024 MPE 2e-6                |     15.67     |     32.77     ||     26.09     |
 | tdnn 7-1024 MPE 2e-5 1.mdl          |     15.54     |     32.77     ||     26.29     |
 | tdnn 7-1024 MPE 1e-5 4.mdl          |     15.76     |     33.55     ||     27.20     |
 | tdnn 7-1024 MPE adapt-PABX          |     14.80     |     30.48     ||     22.56     |
 -----------------------------------------------------------------------------------------
 | spn 7-1024 xEnt                     |     16.49     |     36.23     ||     24.59     |
 | spn 7-1024 xEnt xEnt-PA_user 101.mdl|     16.19     |     33.22     ||     22.69     |
 | spn 7-1024 xEnt xEnt-PA_user mpe    |     15.24     |     32.77     ||     21.65     |
 | spn 7-1024 MPE-1000H 23.mdl         |     15.29     |     33.09     ||     21.65     |
 | spn 7-1024 MPE adapt-PA_all 29.mdl  |     15.11     |     33.42     ||     21.84     |
 | spn 7-1024 MPE adapt-PA_user 2e-5   |     15.31     |     31.79     ||     20.14     |
 | spn 7-1024 MPE adapt-PA_user Hs 2e-5|     15.32     |     32.24     ||     20.93     |
 =========================================================================================

 =====================================================================================================
 |     LM / config                           |     KeHu      |KeHu check_zxm_recheck|  KeHu final   |
 -----------------------------------------------------------------------------------------------------
 |     baseline                              |     20.14     |         19.40        |     18.26     |
 -----------------------------------------------------------------------------------------------------
 |bank+baoxian.chart+word.w0.9               |       -       |         19.27        |     17.88     |
 |bank+baoxian+guojiadianwang.chart+word.w0.9|       -       |         19.02        |     17.94     |
 |bank+baoxian+guojiadianwang_w0.9           |       -       |         19.02        |       -       |
 |bank+baoxian_w0.9                          |       -       |         19.20        |       -       |
 |baoxian+bank_w0.9                          |       -       |         19.02        |       -       |
 |baoxian+user200h.chart+word.w0.9           |       -       |         19.20        |       -       |
 |baoxian+user200h_w0.8                      |       -       |         19.02        |       -       |
 |baoxian+user200h_w0.9                      |       -       |         19.08        |     18.01     |
 |baoxian+user200h.w0.9w0.9                  |       -       |         19.08        |       -       |
 |baoxian+user200h.chart1e-7.w0.9w0.1        |       -       |           -          |     23.82     |
 |baoxian+user200h.chart1e-7.w0.9w0.9        |       -       |           -          |     18.26     |
 =====================================================================================================


  • LiaoNingYiDong: 
 =========================================================================
 |     AM / config                     |     LNYD      |  LNYD re-tag  |
 -------------------------------------------------------------------------
 | tdnn 7-2048 xEnt                    |     21.51     |               |
 | tdnn 7-2048 MPE                     |     20.09     |               |
 | tdnn 7-2048 MPE adapt-LNYD          |     17.92     |     16.29     |
 -------------------------------------------------------------------------
 | tdnn 7-1024 xEnt                    |     21.72     |               |
 | tdnn 7-1024 MPE                     |     20.99     |               |
 | cnn 7-1024 xEnt 600.mdl             |     21.03     |               |
 | cnn 7-1024 MPE 12.mdl               |     19.80     |               |
 | cnn 7-1024 MPE adapt-LNYD 41.mdl    |     17.96     |     15.93     |
 -------------------------------------------------------------------------
 | spn 7-1024 xEnt                     |     21.70     |               |
 | spn 7-1024 MPE-1000H 23.mdl         |     19.97     |               |
 | spn 7-1024 MPE adapt-LNYD           |     18.67     |               |
 | spn cnn 7-1024 xEnt 300.mdl         |     22.26     |               |
 ========================================================================

Embedding

  • The size of nnet1 AM is 6.4M (3M after decomposition). So we need to control AM size within 10M.
  • 5*500-2400 TDNN no-svd/svd100 model, MPE training done 
LM=1e-5, beam=9, max-active=5000
 =============================================================================================================
 |         AM / testset              |  test_1000ju  |  test_2000ju  |  test_8000ju  |  test_10000ju  |
 -------------------------------------------------------------------------------------------------------------
 | nnet1 4*600+800 xEnt (6.4M)       |     25.30     |     40.48     |               |                |
 | nnet1 4*600+800 mpe  (6.4M)       |     20.75     |     35.33     |               |                |
 -------------------------------------------------------------------------------------------------------------
 | nnet3 5*500 mpe (13M)             |     16.18     |     29.53     |               |                |
 | nnet3 5*500 svd-100 mpe (9.5M)    |     17.69     |     30.11     |               |                |
 =============================================================================================================

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
  • 1e-7(13M) wer17.91 compared with 1e-7(no-boundary,71M) 13.4
  • 1e-8(54M) wer17.54
  • Prepare specific domain vocabulary
  • Dianxin/Baoxian/Dianli
  • DT lm training
  • ReFr
  • Merge Character-LM & word-LM
  • Union
  • Compose, success.
  • 2-step decoding: first, character-based LM. Then, word-based LM.

Problem

  • Pingan & Yueyu too much deletion error.
  • TDNN deletion error rate > DNN deletion error rate
  • TDNN Silence scale is too sensitive for different test cases.
  • cmvn causes performance reduction.

SiaSun Robot

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

SID

Digit

  • Engine Package
取自“http://index.cslt.org/mediawiki/index.php?title=Sinovoice-2016-5-26&oldid=20478”