Chapter 8. A Survey of Discriminative Language Modeling Approaches for Large Vocabulary Continuous Speech Recognition

  1. Joseph Keshet2 and
  2. Samy Bengio3
  1. Brian Roark

Published Online: 14 JAN 2009

DOI: 10.1002/9780470742044.ch8

Automatic Speech and Speaker Recognition: Large Margin and Kernel Methods

Automatic Speech and Speaker Recognition: Large Margin and Kernel Methods

How to Cite

Roark, B. (2009) A Survey of Discriminative Language Modeling Approaches for Large Vocabulary Continuous Speech Recognition, in Automatic Speech and Speaker Recognition: Large Margin and Kernel Methods (eds J. Keshet and S. Bengio), John Wiley & Sons, Ltd, Chichester, UK. doi: 10.1002/9780470742044.ch8

Editor Information

  1. 2

    IDIAP Research Institute, Martigny, Switzerland

  2. 3

    Google Research, Google Inc., Mountain View, CA, USA

Author Information

  1. Center for Spoken Language Understanding, Division of Biomedical Computer Science, Oregon Health & Science University, Portland, Oregon, USA

Publication History

  1. Published Online: 14 JAN 2009
  2. Published Print: 16 JAN 2009

ISBN Information

Print ISBN: 9780470696835

Online ISBN: 9780470742044

SEARCH

Keywords:

  • discriminative language modeling and discriminative language model training;
  • language modeling improvements in large vocabulary continuous speech recognition (LVCSR);
  • LVCSR and ‘acoustic-sensitive’ language model training;
  • linear models - general framework;
  • training data and GEN function;
  • weighted finite-state automata (WFA);
  • cross-validation and significant word-error rate (WER);
  • deterministic automaton reading in symbols and parameter weights;
  • syntactic language models and achieving WER reductions

Summary

This chapter contains sections titled:

  • Introduction

  • General Framework

  • Further Developments

  • Summary and Discussion

  • References