Performance Improvement of Dialogue Speech Recognition Method Using Syntactic Rules and Preterminal Bigrams

Bibliographic Information

Other Title
  • 構文規則と前終端記号バイグラムを併用する対話音声認識手法の高速化と高性能化

Search this article

Description

We are studying a CFG-based speech parser that outputs scored subtree sequences. In order to deal with spontaneous speech, the syntactic rules are written to parse subtrees rather than sentences. Bigrams of preterminal symbols are used during beam search. We previously conducted pause unitbased experiments and have confirmed that our method yields good performance. In this paper, we report on two kinds of performance improvement; i.e., the dictionary access improvement and the improvement of beam search. We also discuss ways to deal with more than one pause unit.

Journal

Citations (1)*help

See more

References(11)*help

See more

Details 詳細情報について

  • CRID
    1573950402205401216
  • NII Article ID
    110003278156
  • NII Book ID
    AN10091225
  • Text Lang
    ja
  • Data Source
    • CiNii Articles

Report a problem

Back to top