“2014-07-18”版本间的差异

2014年7月18日 (五) 02:34的最后版本

Resoruce Building

Leftover questions

• Asymmetric window: Great improvement on training set(WER 34% to 24%), however the improvement is lost on test.
• Multi GPU training: Error encountered
• Multilanguage training
• Investigating LOUDS FST.
• CLG embedded decoder plus online compiler.
• DNN-GMM co-training

AM development

Sparse DNN

• GA-based block sparsity (++++++++++)

Noise training

• Journal paper writing on going

Multilingual ASR

LM = Tel201406.HW.v2.1.1

     AM\testset    |JS27H_100|  JS_2h  |ShanXi_2h|ShaanXi2h|Unknown2h|   ENG   |
 Tel201406.v1.0.S  |         |    -    |    -    |    -    |    -    |    -    |
 Tel201406.v1.1.S  |    -    |    -    |    -    |    -    |    -    |    -    |
Tel201406.HW.v2.0.B|  20.18  |  17.49  |  23.85  |  22.81  |  22.48  |  83.28  |
Tel201406.HW.v2.0.S|  19.95  |  17.74  |  23.73  |  22.36  |  22.49  |  67.70  |
Tel201406.HW.v2.1.B|  19.14  |  16.97  |  24.26  |  22.28  |  22.97  |  85.41  |
Tel201406.HW.v2.1.S|  19.44  |  17.62  |  24.49  |  23.06  |  23.60  |  74.61  |

• v1.*: no English words involved.
• v2.*: with English words involved.

Denoising & Farfield ASR

• Sparse linear prediction. Need to correct the model.
• Use xEnt as the adaptation object, instead of MSE based feature mapping
• Use the simulation tool to add reverberation.
• [1]
• Investigate the impact of speech rate. Use Tencent 200h data to conduct the experiments.
• Investigate the correlation between phone speed & entropy.

VAD

• Waiting for engineering work

Scoring

• Refine the acoustic model with AMIDA database. problem solved by involving both wsj and AMIDA.
• Model ready for picking up

Embedded decoder

• The first deliver is Emb201407_BG_v0.0
• Train two smaller network: 500x4+600, 400x4+500

LM development

Domain specific LM

h2. Domain specific LM construction

h3. TAG LM

• Some problems with the tagging. all numbers are tagged.

h3. Chatting LM

• Building chatting lexicon

Word2Vector

W2V based doc classification

• Initial results variable Bayesian GMM obtained. Performance is not as good as the conventional GMM.

Semantic word tree

• Version v2.0 released (filter with query log)
• Please deliver to /nfs/disk/perm/data/corpora/semanticTree (Xingchao)
• Version v3.0 under going. Further refinement with Baidu Baike hierarchy

NN LM

• Character-based NNLM (6700 chars, 7gram), 500M data training done.

• Inconsistent pattern in WER were found on Tenent test sets
• probably need to use another test set to do investigation.

• Investigate MS RNN LM training

Speaker ID

• reading materials
• prepare to run sre08

Translation

• Initial version released
• collecting more data (Xinhua parallel text, bible, name entity) for the second version