Speaker adaptation for Hidden Marcov models

D. Shaked, A. Cohen

Research output: Contribution to conferencePaperpeer-review

Abstract

This paper proposes an algorithm for adaptive training of a Hidden Markov Model (HMM) in an Automatic Speech Recognition (ASR) System. The adaptive training procedure uses the Viterbi algorithm. A forgeting factor enables regulation of the model's adaptation dynamics as well as the steady state stability, according to a planed adaptation policy. The proposed algorithm was tested by synthesized data, generated by HMMs similar to models used for in speech recognition. Results of the simulation using the divergence measure, illustrate the effect of various adaptation policies, on the adaptation time and accuracy.

Original languageEnglish
DOIs
StatePublished - 1 Jan 1989
Event16th Conference of Electrical and Electronics Engineers in Israel, EEIS 1989 - Tel-Aviv, Israel
Duration: 7 Mar 19899 Mar 1989

Conference

Conference16th Conference of Electrical and Electronics Engineers in Israel, EEIS 1989
Country/TerritoryIsrael
CityTel-Aviv
Period7/03/899/03/89

ASJC Scopus subject areas

  • Software
  • Computer Networks and Communications
  • Signal Processing
  • Electrical and Electronic Engineering
  • Instrumentation

Fingerprint

Dive into the research topics of 'Speaker adaptation for Hidden Marcov models'. Together they form a unique fingerprint.

Cite this