Sinovoice-2016-4-28

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

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
  -----------------------------------------------------------------------------

• 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 | 9.13/481
  -----------------------------------------------------------------------------

• 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
  -----------------------------------------------------------------------------

• 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
  -----------------------------------------------------------------------------

• Add one silence arc from start-state to end-state

Big-Model Training

• 16k

 ================================================================================================
 |                      |   TDNN 7-1200   | TDNN 7-1200 enhance | TDNN 7-1200 svd600 |
 ------------------------------------------------------------------------------------------------
 |8000ju frame_skip=1   |                 |   0.0556 / 0.349    |  0.0559 / 0.306    |
 |8000ju frame_skip=2   |  0.059 / 0.243  |   0.0591 / 0.231    |  0.0589 / 0.228    |
 ------------------------------------------------------------------------------------------------
 |10000ju frame_skip=1  |                 |   0.1241 / 0.341    |  0.1244 / 0.358    |
 |10000ju frame_skip=2  |  0.1348 / 0.234 |   0.1315 / 0.245    |  0.1311 / 0.204    |
 ------------------------------------------------------------------------------------------------
 |English frame_skip=1  |                 |   0.3897 / 0.370    |  0.4062 / 0.353    |
 |English frame_skip=2  |  0.4296         |   0.4237 / 0.276    |  0.4306 / 0.252    |
 ================================================================================================

• 8k

 PingAn:
 ===============================================================================
 |     AM / config            |   all beam9   |all beam9 biglm||  KeHu beam9   |
 -------------------------------------------------------------------------------
 | tdnn 7-2048 xEnt           |     16.45     |     16.22     || 36.49 / 25.18 |
 | tdnn 7-2048 MPE            |     15.22     |     14.87     || 32.77 / 23.48 |
 | tdnn 7-2048 MPE adapt-PABX |     14.67     |     14.63     || 31.33 / 22.76 |
 -------------------------------------------------------------------------------
 | tdnn 7-1024 xEnt           |     16.60     |     16.25     || 35.91 / 25.58 |
 | tdnn 7-1024 MPE            |     15.67     |     15.61     || 32.77 / 26.09 |
 | tdnn 7-1024 MPE adapt-PABX |     14.80     |     14.76     || 30.48 / 22.56 |
 ===============================================================================

 LiaoNingYiDong:
 ==============================================================================
 |     AM / config            |     beam9     |   beam9 biglm |     beam13    |
 ------------------------------------------------------------------------------
 | tdnn 7-2048 xEnt           |     21.51     |     21.05     |     21.17     |
 | tdnn 7-2048 MPE            |     20.09     |     19.74     |     19.74     |
 | tdnn 7-2048 MPE adapt-LNYD |     17.92     |     17.87     |     17.58     |
 ------------------------------------------------------------------------------
 | tdnn 7-1024 xEnt           |     21.72     |     22.74     |     21.64     |
 | tdnn 7-1024 MPE            |     20.99     |     20.77     |     20.74     |
 | tdnn 7-1024 MPE adapt-LNYD |               |               |               |
 ==============================================================================

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.

SinSong Robot

• Test based on 10000h(7*2048-xent) model

 ------------------------------------------------
   condition | clean  | replay(0.5m) | real-env
 ------------------------------------------------
     wer     |   3    |  18(mpe-14)  | too-bad
 ------------------------------------------------

• Plan to record in restaurant on April 10.

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

• Union
• Compose, success.

• 2-step decoding: first, character-based LM. Then, word-based LM.
• Word boundary character training

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

• DNN-PLDA gets better performance than i-Vector;

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