Presentation	1993/8/19 Study of Training Section for Concatenated Training of Phoneme HMM Shigeru Homma,
PDF Download Page	PDF download Page Link
Abstract(in Japanese)	(See Japanese page)
Abstract(in English)	This paper is concerned with restriction of training section for concatenated training of phoneme HMM(Hidden Markov Model).The training section used for training concatenated phoneme HMM is whole speech data when conventional method is used.In this case,a large surplus of data is also used to re-estimate the model parameters,and,this may be the reason for poor recognition rate. Therefore restriction of training section is necessary to get preferable model parameters.For speaker-dependent phoneme recognition in Japanese continuous speech when whole context, excluding object phoneme,is known,the relationship between extent of training section and recognition rate is examined.By restriction of training section based on pause or hand-marking, same recognition rate as conventional training is obtained after about 1, 3 iteration number of training.Recognition error is reduced 60% at best.
Keyword(in Japanese)	(See Japanese page)
Keyword(in English)	restriction of training section / phoneme HMM / concatenated training
Paper #	SP93-51
Date of Issue

Conference Information
Committee	SP
Conference Date	1993/8/19(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	Speech (SP)
Language	JPN
Title (in Japanese)	(See Japanese page)
Sub Title (in Japanese)	(See Japanese page)
Title (in English)	Study of Training Section for Concatenated Training of Phoneme HMM
Sub Title (in English)
Keyword(1)	restriction of training section
Keyword(2)	phoneme HMM
Keyword(3)	concatenated training
1st Author's Name	Shigeru Homma
1st Author's Affiliation	NTT Human Interface Laboratories()
Date	1993/8/19
Paper #	SP93-51
Volume (vol)	vol.93
Number (no)	183
Page	pp.pp.-
#Pages	8
Date of Issue