HMM算法在语音识别中的应用——算法学习

总体框架

输入

Treat acoustic(听觉的) input O as sequence of individual observations

O=o1,o2,...,ot

结果

Define sentence as a sequence of words

W=w1,w2,...,wn

判断模型

• 最大概率： W=arg maxW∈L P(W|O)
• 贝叶斯： W=arg maxW∈L P(O|W)P(W)P(O)
• 化简：由于 P(O) 对所有 W 一样， W=arg maxW∈L P(O|W)P(W)

模型

• Feature Extraction: 39 “MFCC” features
• Acoustic Model: Gaussians for computing p(o|q)

• Lexicon(词典)/Pronunciation(发音) Model: HMM, what phones can follow each other
  

  ---

  

• Language Model: N-grams for computing p(wi|wi−1)

markov chian

• states: Q=q1,q2,...,qN , qt is the state at time t.
• transition probability: A=[a11,a12,...,aNN]
  
  • aij is the probability of trasition from i to j.
  • aij=P(qt−1=i|qt=j)
  • ∑Nj=1aij=1 , for i∈[1,N]
• markov assumption:
  
  • P(qi|q1q2...qi−1)=P(qi|qi−1)
• initial status
  
  • πi=P(q1=i)
  • ∑Nj=1πj=1

hidden markov chian

three problem

thrid problem

以上所有内容从以下附件中抽取

• lec3.pdf
• 斯坦福大学关于语音识别的全部课件