Presentation 1997/1/16
Speaker adaptation based on Tree-Structured Tied-Difference vectors and Their Confidence.
Kazuki Shibata, Hiroshi Matsumoto,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In speaker adaptation based on the Tree-Structured probability density function (pdf) of HMMs, this paper propose a method to improve adaptation accuracy using a confidence parameter in maximum a posteriori (MAP) estimation. In this method, the directional components of difference vectors for adaptation are tied together, and their magnitudes are independently estimated by an approximate MAP method. The confidence parameter of a prior model is optimized depending on the number of training samples assigned to a cluster of pdfs. Phoneme recognition tests show that the proposed method attaines a 2.4 to 2.9% higher recognition accuracy than the original tree-structured-based method.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Speaker Adaptation / MAP Estimation / Continuous HMM / Tree-Structure / Speech Recognition
Paper # SP96-90
Date of Issue

Conference Information
Committee SP
Conference Date 1997/1/16(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) Speaker adaptation based on Tree-Structured Tied-Difference vectors and Their Confidence.
Sub Title (in English)
Keyword(1) Speaker Adaptation
Keyword(2) MAP Estimation
Keyword(3) Continuous HMM
Keyword(4) Tree-Structure
Keyword(5) Speech Recognition
1st Author's Name Kazuki Shibata
1st Author's Affiliation Faculty of Engineering, Shinshu University()
2nd Author's Name Hiroshi Matsumoto
2nd Author's Affiliation Faculty of Engineering, Shinshu University
Date 1997/1/16
Paper # SP96-90
Volume (vol) vol.96
Number (no) 448
Page pp.pp.-
#Pages 7
Date of Issue