Presentation 2009-12-21
Experimental study of acoustic modeling using speaker-invariant speech contrast as modeling unit
Daisuke SAITO, Ryo MATSUURA, Nobuaki MINEMATSU, Keikichi HIROSE,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Speech acoustics vary due to differences in age, gender, vocal tract length, microphone, and so on. The authors recently proposed a structural and abstract representation of speech, where these variations were effectively removed. This representation captures only dynamics of speech. In our previous studies, using this abstract representation, an ASR framework, which we call the structure-based ASR, was proposed and examined. However, there are two problems for the structure-based ASR; the curse of dimensionality and the large size of modeling unit. As a solution for these problems, this report proposes a new acoustic modeling based on parameter sharing of statistical structure models and efficient reuse of the shared models. In the proposed method, edge vectors, which represent speech contrasts between any two acoustic events, are considered and parameter sharing is carried out based on clustering in the parametric space of edge vectors. To construct an acoustic model for a new word, the most likely edge models are selected and allocated for each edge vector in the structure of that new word. Experiments of recognition using continuous utterances of Japanese vowels show the validity of the proposed method.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) structural representation / invariant features / speech contrasts / clustering / acoustic modeling
Paper # NLC2009-13,SP2009-77
Date of Issue

Conference Information
Committee NLC
Conference Date 2009/12/14(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) Experimental study of acoustic modeling using speaker-invariant speech contrast as modeling unit
Sub Title (in English)
Keyword(1) structural representation
Keyword(2) invariant features
Keyword(3) speech contrasts
Keyword(4) clustering
Keyword(5) acoustic modeling
1st Author's Name Daisuke SAITO
1st Author's Affiliation Graduate School of Engineering, The University of Tokyo()
2nd Author's Name Ryo MATSUURA
2nd Author's Affiliation Graduate School of Frontier Sciences, The University of Tokyo
3rd Author's Name Nobuaki MINEMATSU
3rd Author's Affiliation Graduate School of Information Science and Technology, The University of Tokyo
4th Author's Name Keikichi HIROSE
4th Author's Affiliation Graduate School of Information Science and Technology, The University of Tokyo
Date 2009-12-21
Paper # NLC2009-13,SP2009-77
Volume (vol) vol.109
Number (no) 355
Page pp.pp.-
#Pages 6
Date of Issue