“ASR:2015-07-13”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
Text Group Intern Project
Zxw讨论 | 贡献
Speech Processing
 
第2行: 第2行:
 
=== AM development ===
 
=== AM development ===
  
==== Environment ====*
+
==== Environment ====
 
* the GPU of grid-14 does not work
 
* the GPU of grid-14 does not work
  
第16行: 第16行:
 
====Data selection unsupervised learning
 
====Data selection unsupervised learning
 
* acoustic feature based submodular using Pinan dataset --zhiyong
 
* acoustic feature based submodular using Pinan dataset --zhiyong
 +
* write code to speed up --zhiyong
  
  
第36行: 第37行:
  
 
===Dark knowledge===
 
===Dark knowledge===
* test random last output layer when train MPE --zhiyuan
+
* test random last output layer when train MPE --zhiyuan,mengyuan
  
  
第44行: 第45行:
  
 
===rectifier===
 
===rectifier===
 +
* hold
 
* WER performs worse using auraro4 --zhiyuan
 
* WER performs worse using auraro4 --zhiyuan
 
* train using other dataset
 
* train using other dataset

2015年7月16日 (四) 05:40的最后版本

Speech Processing

AM development

Environment

  • the GPU of grid-14 does not work

RNN AM

  • hold
  • morpheme RNN --zhiyuan
  • train using large dataset--mengyuan

Mic-Array

  • hold
  • compute EER with kaldi

====Data selection unsupervised learning

  • acoustic feature based submodular using Pinan dataset --zhiyong
  • write code to speed up --zhiyong


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

  • hold
  • deliver to mengyuan

Speaker ID

  • DNN-based sid --Lantian


Ivector&Dvector based ASR

  • hold --Tian Lan
  • Cluster the speakers to speaker-classes, then using the distance or the posterior-probability as the metric
  • dark-konowlege using i-vector
  • train on wsj(testbase dev93+evl92)
  • --hold

Dark knowledge

  • test random last output layer when train MPE --zhiyuan,mengyuan


language vector

  • train using language vector with the dataset of 1400h_CN + 100h_EN--mengyuan
  • write a paper--zhiyuan

rectifier

  • hold
  • WER performs worse using auraro4 --zhiyuan
  • train using other dataset
  • rectifier RNN

audio embedding=

  • audio ebedding --Wei Xu

Text Processing

RNN LM

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

Neural Based Document Classification

  • (hold)

Order representation

  • Nested Dropout
  • semi-linear --> neural based auto-encoder.
  • modify the objective function(hold)

Balance Representation

  • Find error signal

Recommendation

  • Reproduce baseline.
  • LDA matrix dissovle.
  • LDA (Text classification & Recommendation System) --> AAAI

DSSM based QA

  • Demo Release.

Seq to Seq(09-15)

  • Review papers.(Reported in 07-08)
  • Reproduce baseline.

Text Group Intern Project

  • ====Buddhist Process====

(hold)

RNN Poem Process

(hold)

RNN Document Vector

(hold)

Image Baseline

  • Demo Release.
  • Paper Report.