Discriminative Learning for Speech Recognition: Theory and Practice

  • Li Deng ,
  • Xiaodong He

Published by Morgan & Claypool | October 2008

Publication | Publication | Publication | Publication | Publication | Publication | Publication | Publication | Publication | Publication

This book starts by providing an introduction to discriminative learning, speech recognition, and the roles of discriminative learning in speech recognition. Then it presents some background material on basic probability distributions and on optimization techniques; both serve as mathematical requisites for the remaining book content dealing with detailed techniques for discriminative learning in speech recognition. The basic probability distributions covered in the background material include multinomial distribution and multivariate Gaussian distribution, both belonging to the more general exponential-distribution family, as well as Gaussian mixture distribution, which is outside of the exponential-distribution family. The optimization concepts and techniques covered in the background material include definitions of global and local optimums, their necessary condition, Lagrange multiplier method, gradient-based method, and, fi nally, growth transformation (GT) method.