Presentation 2007/12/13
Speaker Selection for Unsupervised Speaker Adaptation based on HMM Sufficient Statistics
Masahiro TANI, Tadashi EMORI, Yoshifumi OHNISHI, Takafumi KOSHINAKA, Koichi SHINODA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) We propose a new speaker selection method for the unsupervised speaker adaptation based on HMM sufficient statistics. The adaptation technique of using HMM sufficient statistics has been proposed as one of the rapid unsupervised speaker adaptation techniques in speech recognition. The procedure is as follows : First the training speakers acoustically close to the test speaker are selected. Then, the acoustic model is trained using the HMM sufficient statistics of these selected training speakers. In this technique, the number of selected training speakers is always constant. In our proposed speaker selection method, the number of speakers is determined by the distances between the test speaker and each training speaker. In our recognition experiments using spoken dialogue data, the proposed method improved word accuracy by 0.74 points. It was confirmed that the proposed method particularly effective when there are not many training speakers around the test speaker in acoustic space.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Unsupervised Adaptation / HMM Sufficient Statistics / Speaker Selection
Paper # NLC2007-47,SP2007-110
Date of Issue

Conference Information
Committee NLC
Conference Date 2007/12/13(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) Speaker Selection for Unsupervised Speaker Adaptation based on HMM Sufficient Statistics
Sub Title (in English)
Keyword(1) Unsupervised Adaptation
Keyword(2) HMM Sufficient Statistics
Keyword(3) Speaker Selection
1st Author's Name Masahiro TANI
1st Author's Affiliation Central Research Laboratories, NEC Corporation()
2nd Author's Name Tadashi EMORI
2nd Author's Affiliation Central Research Laboratories, NEC Corporation
3rd Author's Name Yoshifumi OHNISHI
3rd Author's Affiliation Central Research Laboratories, NEC Corporation
4th Author's Name Takafumi KOSHINAKA
4th Author's Affiliation Central Research Laboratories, NEC Corporation
5th Author's Name Koichi SHINODA
5th Author's Affiliation Graduate School of Information Science and Engineering, Tokyo Institute of Technology
Date 2007/12/13
Paper # NLC2007-47,SP2007-110
Volume (vol) vol.107
Number (no) 405
Page pp.pp.-
#Pages 5
Date of Issue