“ASR:2015-01-19”版本间的差异

来自cslt Wiki
跳转至： 导航、 搜索
（以“==Speech Processing == === AM development === ==== Environment ==== * May gpu760 of grid-14 be something wrong. To be exchanged. * grid-2 can not be booted * grid-1...”为内容创建页面）
 
Lr（讨论 | 贡献）
（→‎Text Processing）
第81行： 第81行：
 
:* mix the sougou2T-lm,kn-discount continue
 
:* mix the sougou2T-lm,kn-discount continue
 
:* train a large lm using 25w-dict.(hanzhenglong/wxx)
 
:* train a large lm using 25w-dict.(hanzhenglong/wxx)
−
::* data pre-processing("this week")
+
::* find the problem in asr result
−
 
+
::* the model will finish (Tuesday)
−
* new dict.
+
−
:* dongxu help zhenglong with large dictionary.
+
  
 
====tag LM====
 
====tag LM====
 
* Tag Lm
 
* Tag Lm
 
:* tag Probability should test add the weight(hanzhenglong) and handover to hanzhenglong ("this month")
 
:* tag Probability should test add the weight(hanzhenglong) and handover to hanzhenglong ("this month")
 +
:* run a tag demo('''this week''')
 
*paper
 
*paper
 
:* paper submit this week.
 
:* paper submit this week.
第94行： 第93行：
 
:* find similarity word using word2vec,word vector is training.
 
:* find similarity word using word2vec,word vector is training.
 
:* set the weight for word
 
:* set the weight for word
 +
:* set a proper test set
 +
:* write a draft of a paper
  
 
====RNN LM====
 
====RNN LM====
第109行： 第110行：
 
* Knowledge vector  
 
* Knowledge vector  
 
:* Make a proper test set.
 
:* Make a proper test set.
−
:*  
+
:* use text information and train word vector together.
 
:* Modify the object function and training process.
 
:* Modify the object function and training process.
−
 
+
:* try to train on the whole data set
−
====relation====
+
* result
 +
:* 0.745->0.79, using yago for training.
 
====Character to word====
 
====Character to word====
 
* Character to word conversion(hold)
 
* Character to word conversion(hold)
第119行： 第121行：
 
* v5.0 demo released
 
* v5.0 demo released
 
:* cut the dict and use new segment-tool
 
:* cut the dict and use new segment-tool
 +
 +
===Sparse NN in NLP===
 +
* review related paper
  
 
===QA===
 
===QA===
第126行： 第131行：
 
:* add more feature to improve search.
 
:* add more feature to improve search.
 
::* POS, NER ,tf ,idf ..
 
::* POS, NER ,tf ,idf ..
 +
:* extract more features about lexical, syntactic and semantic to improve re-ranking performance.
  
 
====context framework====
 
====context framework====
−
*  code for organization  
+
*  code for organization
 +
::* change to knowledge graph
 
====query normalization====
 
====query normalization====
 
* using NER to normalize the word
 
* using NER to normalize the word
  
 
* new inter will install SEMPRE
 
* new inter will install SEMPRE

2015年1月23日 (五) 01:23的版本

Speech Processing

AM development

Environment

  • May gpu760 of grid-14 be something wrong. To be exchanged.
  • grid-2 can not be booted
  • grid-10 should replac the CPU fan.

Sparse DNN

RNN AM

Dropout & Maxout & retifier

  • Drop out
  • MaxOut && P-norm(+)
  • Need to solve the too small learning-rate problem
    • Add one normalization layer after the pnorm-layer
    • Add L2-norm upper bound
  • hold

Convolutive network

  • Convolutive network(DAE)
  • Feature extractor
    • Combined with raw features, better performance obsearved.
    • Technical report to draft, Yiye Lin, Shi Yin, Menyuan Zhao and Mian Wang(+)
  • To test real enviroment echo.(+)

DNN-DAE(Deep Atuo-Encode-DNN)

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

VAD

  • Harmonics and Teager energy features done.
  • Frame energy Model training done.
  • Other features model to be train.

Speech rate training

Confidence

  • Reproduce the experiments on fisher dataset.
  • Use the fisher DNN model to decode all-wsj dataset
  • preparing scoring for puqiang data
  • HOLD

Neural network visulization

Speaker ID

Language ID

Voice Conversion

  • Yiye is reading materials
  • HOLD


Text Processing

LM development

Domain specific LM

  • LM2.1
  • mix the sougou2T-lm,kn-discount continue
  • train a large lm using 25w-dict.(hanzhenglong/wxx)
  • find the problem in asr result
  • the model will finish (Tuesday)

tag LM

  • Tag Lm
  • tag Probability should test add the weight(hanzhenglong) and handover to hanzhenglong ("this month")
  • run a tag demo(this week)
  • paper
  • paper submit this week.
  • similar word extension in FST
  • find similarity word using word2vec,word vector is training.
  • set the weight for word
  • set a proper test set
  • write a draft of a paper

RNN LM

  • rnn
  • test wer RNNLM on Chinese data from jietong-data
  • generate the ngram model from rnnlm and test the ppl with different size txt.
  • lstm+rnn
  • check the lstm-rnnlm code about how to Initialize and update learning rate.(hold)

Word2Vector

W2V based doc classification

  • data prepare.

Knowledge vector

  • Knowledge vector
  • Make a proper test set.
  • use text information and train word vector together.
  • Modify the object function and training process.
  • try to train on the whole data set
  • result
  • 0.745->0.79, using yago for training.

Character to word

  • Character to word conversion(hold)

Translation

  • v5.0 demo released
  • cut the dict and use new segment-tool

Sparse NN in NLP

  • review related paper

QA

improve fuzzy match

  • add Synonyms similarity using MERT-4 method(hold)

improve lucene search

  • add more feature to improve search.
  • POS, NER ,tf ,idf ..
  • extract more features about lexical, syntactic and semantic to improve re-ranking performance.

context framework

  • code for organization
  • change to knowledge graph

query normalization

  • using NER to normalize the word
  • new inter will install SEMPRE
取自“http://index.cslt.org/mediawiki/index.php?title=ASR:2015-01-19&oldid=13635”