Presentation 1996/12/13
Acoustic Model Generation Using State Clustering by Infomation Criterion
Koichi SHINODA, Takao WATANABE,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In recent years, context-dependent subword units have been often employed for large-vocabulary speech recognition systems using HMMs. Since the amount of training data available is usually not large enough to estimate the parameters of all the units with sufficient accuracy, most of these systems use clustering methods to reduce the degree of freedom. However, none of these clustering methods have given a stop criterion of the clustering. In this paper, we propose one clustering method which use MDL criterion as the stop criterion. The cvaluation experiment proved that the proposed method achieved as high recognition accuracy as conventional heuristic methods with less computational amount.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) speech recognition / Hidden Markov Model / recognition unit / information criteria / Minimum Descritption Length Principle
Paper # NLC96-48,SP96-79
Date of Issue

Conference Information
Committee NLC
Conference Date 1996/12/13(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Natural Language Understanding and Models of Communication (NLC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Acoustic Model Generation Using State Clustering by Infomation Criterion
Sub Title (in English)
Keyword(1) speech recognition
Keyword(2) Hidden Markov Model
Keyword(3) recognition unit
Keyword(4) information criteria
Keyword(5) Minimum Descritption Length Principle
1st Author's Name Koichi SHINODA
1st Author's Affiliation NEC Information Technology Research Laboratories()
2nd Author's Name Takao WATANABE
2nd Author's Affiliation NEC Information Technology Research Laboratories
Date 1996/12/13
Paper # NLC96-48,SP96-79
Volume (vol) vol.96
Number (no) 420
Page pp.pp.-
#Pages 7
Date of Issue