“ASR:2015-05-11”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
Text Processing
Lr讨论 | 贡献
RNN LM
 
(相同用户的5个中间修订版本未显示)
第49行: 第49行:
 
==Text Processing==
 
==Text Processing==
 
====RNN LM====
 
====RNN LM====
*rnn
+
*character-lm rnn(hold)
:* test the ppl and code the character-lm(hold)
+
 
*lstm+rnn
 
*lstm+rnn
 
:* check the lstm-rnnlm code about how to Initialize and update learning rate.(hold)
 
:* check the lstm-rnnlm code about how to Initialize and update learning rate.(hold)
第59行: 第58行:
 
===Translation===
 
===Translation===
 
* similar-pair method in English word using translation model.
 
* similar-pair method in English word using translation model.
 +
:* result:wer:70%-50% on top1.
 +
:* change the AM model
  
 
===Order representation ===
 
===Order representation ===
 
* modify the objective function
 
* modify the objective function
 
* sup-sampling method to solve the low frequence word
 
* sup-sampling method to solve the low frequence word
* learn binary vector
+
===binary vector===
  
 
===Stochastic ListNet===
 
===Stochastic ListNet===
* using sampling method
+
* using sampling method and test
  
 
===relation classifier===
 
===relation classifier===
* majority of error in others class
+
* test the  bidirectional neural network(B-RNN) and get a little improvement
*
+
 
 +
===plan to do===
 +
* combine LDA with neural network

2015年5月18日 (一) 01:24的最后版本

Speech Processing

AM development

Environment

  • grid-15 often does not work

RNN AM

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
  • compute EER with kaldi

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

Speaker ID

Ivector&Dvector based ASR

  • hold --Tian Lan

Dark knowledge

  • Ensemble using 100h dataset to construct diffrernt structures -- Mengyuan
  • 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

Text Processing

RNN LM

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

W2V based document classification

  • make a technical report about document classification using CNN --yiqiao
  • CNN adapt to resolve the low resource problem

Translation

  • similar-pair method in English word using translation model.
  • result:wer:70%-50% on top1.
  • change the AM model

Order representation

  • modify the objective function
  • sup-sampling method to solve the low frequence word

binary vector

Stochastic ListNet

  • using sampling method and test

relation classifier

  • test the bidirectional neural network(B-RNN) and get a little improvement

plan to do

  • combine LDA with neural network