ASR:2015-04-20

Speech Processing

AM development

Environment

• grid-11 often shut down automatically, too slow computation speed.
• New grid-13 added, using gpu970
• To update the wiki enviroment infomation

RNN AM

• details at http://liuc.cslt.org/pages/rnnam.html
• Test monophone on RNN using dark-knowledge
• run using wsj,MPE

Mic-Array

• Change the prediction from fbank to spectrum features
• investigate alpha parameter in time domian and frquency domain
• ALPHA>=0, using data generated by reverber toolkit
• consider theta

RNN-DAE(Deep based Auto-Encode-RNN)

• HOLD --Zhiyong Zhang
• http://cslt.riit.tsinghua.edu.cn/cgi-bin/cvss/cvss_request.pl?account=zhangzy&step=view_request&cvssid=261

Speaker ID

• DNN-based sid --Yiye Lin
• http://cslt.riit.tsinghua.edu.cn/cgi-bin/cvss/cvss_request.pl?account=zhangzy&step=view_request&cvssid=327

Ivector&Dvector based ASR

• Cluster the speakers to speaker-classes, then using the distance or the posterior-probability as the metric
• Direct using the dark-knowledge strategy to do the ivector training.
• http://cslt.riit.tsinghua.edu.cn/cgi-bin/cvss/cvss_request.pl?step=view_request&cvssid=340
• Ivector dimention is smaller, performance is better
• Augument to hidden layer is better than input layer
• train on wsj(testbase dev93+evl92)

Dark knowledge

• Ensemble

  http://cslt.riit.tsinghua.edu.cn/cgi-bin/cvss/cvss_request.pl?account=zxw&step=view_request&cvssid=264 --Zhiyong Zhang

• adaptation for chinglish under investigation -- Mengyuan Zhao

• Try to improve the chinglish performance extremly

• unsupervised training with wsj contributes to aurora4 model --Xiangyu Zeng

• test large database with AMIDA

bilingual recognition

• http://cslt.riit.tsinghua.edu.cn/cgi-bin/cvss/cvss_request.pl?account=zxw&step=view_request&cvssid=359 --Zhiyuan Tang

Text Processing

tag LM

• similar word extension in FST

• will check the formula using Bayes and experiment
• add similarity weight

RNN LM

• rnn

• test the ppl and code the character-lm

• lstm+rnn

• check the lstm-rnnlm code about how to Initialize and update learning rate.(hold)

W2V based document classification

• result about norm model [1]
• try CNN model

Translation

• v5.0 demo released

• cut the dict and use new segment-tool

Sparse NN in NLP

• test the drop-out model and the performance gets a little improvement, need some result:
• test the order feature ,need some result:
• large dimension result:http://cslt.riit.tsinghua.edu.cn/cgi-bin/cvss/cvss_request.pl?account=lr&step=view_request&cvssid=344

• sparse-nn on 1000 dimension(le-6,0.705236) is better than 200 dimension(le-12,0.694678).

online learning

• modified the listNet SGD

relation classifier

• check the CNN code and contact the author of paper