SPEAKER-INDEPENDENT CONNECTED WORDS RECOGNITION USING VECTOR QUANTIZATION OF THE TIME-SPACE PATTERN.

Satoru Hayamizu, Ryu ichi Oka

Research output: Contribution to journalArticlepeer-review

Abstract

This paper describes an attempt on the word recognition using a matrix of feature vectors as a description unit for a time pattern. Pseudo-phoneme patterns represented by the matrices are extracted in advance by a vector quantization. The connected words recognition is conducted using these pseudo-phoneme patterns and Continuous Dynamic Programming. Distance measures among matrices and an algorithm of the vector quantization are described. Results of preparatory experiment are shown for 54 city names.

Original languageEnglish
Pages (from-to)160-163
Number of pages4
JournalDenshi Gijutsu Sogo Kenkyusho Iho/Bulletin of the Electrotechnical Laboratory
Volume50
Issue number2-3
Publication statusPublished - 1986
Externally publishedYes

ASJC Scopus subject areas

  • Condensed Matter Physics
  • Electrical and Electronic Engineering

Fingerprint Dive into the research topics of 'SPEAKER-INDEPENDENT CONNECTED WORDS RECOGNITION USING VECTOR QUANTIZATION OF THE TIME-SPACE PATTERN.'. Together they form a unique fingerprint.

Cite this