ASR:2015-04-08

目录 [隐藏] 1 Speech Processing 1.1 AM development 1.1.1 Environment 1.1.2 RNN AM 1.1.3 Mic-Array 1.1.4 Convolutive network 1.1.5 RNN-DAE(Deep based Auto-Encode-RNN) 1.2 Speaker ID 1.3 Ivector based ASR 2 Text Processing 2.1 tag LM 2.1.1 RNN LM 2.1.2 W2V based doc classification 2.2 Translation 2.3 Sparse NN in NLP 2.4 online learning Speech Processing[编辑] AM development[编辑] Environment[编辑] grid-11 often shut down automatically, too slow computation speed.

RNN AM[编辑] details at http://liuc.cslt.org/pages/rnnam.html tuning parameters on monophone NN run using wsj,MPE

Mic-Array[编辑] investigate alpha parameter in time domian and frquency domain ALPHA>=0

Convolutive network[编辑] HOLD CNN + DNN feature fusion RNN-DAE(Deep based Auto-Encode-RNN)[编辑] HOLD -Zhiyong http://cslt.riit.tsinghua.edu.cn/cgi-bin/cvss/cvss_request.pl?account=zhangzy&step=view_request&cvssid=261

Speaker ID[编辑] DNN-based sid --Yiye Decode --Yiye http://cslt.riit.tsinghua.edu.cn/cgi-bin/cvss/cvss_request.pl?account=zhangzy&step=view_request&cvssid=327 Ivector based ASR[编辑] 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) Text Processing[编辑] tag LM[编辑] similar word extension in FST check the formula using Bayes and experiment RNN LM[编辑] rnn code the character-lm using Theano lstm+rnn check the lstm-rnnlm code about how to Initialize and update learning rate.(hold) W2V based doc classification[编辑] corpus ready learn some benchmark. Translation[编辑] v5.0 demo released cut the dict and use new segment-tool Sparse NN in NLP[编辑] prepare the ACL check the code to find the problem . increase the dimension use different test set,but the result is not good. online learning[编辑] data is ready.prepare the ACL paper prepare sougouQ data and test it using current online learning method baseline is not normal.