Presentation 2012-11-08
Improvements of HMM-based Speech Synthesis Using Rich Context Models
Shinnosuke TAKAMICHI, Tomoki TODA, Yoshinori SHIGA, Sakriani SAKTI, Graham NEUBIG, Satoshi NAKAMURA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In the traditional HMM-based speech synthesis, generated speech parameters tend to be excessively smoothed. To alleviate this problem, we have proposed a parameter generation method with rich context models in our previous work. This method improves speech quality while keeping the flexibility of HMM-based speech synthesis. However, synthetic speech still sounds muffled because the generated parameters strongly depend on over-smoothed initial parameters in iterative parameter generation procedure. In this paper, we propose an initialization method for generating less-smoothed initial parameters using context-clustered HMMs based on a large-sized decision tree. Experimental evaluations of the proposed method demonstrate that the proposed method yields significant improvements in the quality of synthetic speech.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) HMM-based speech synthesis / rich context model / parameter generation / tree-based context clustering
Paper # SP2012-78
Date of Issue

Conference Information
Committee SP
Conference Date 2012/11/1(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) Improvements of HMM-based Speech Synthesis Using Rich Context Models
Sub Title (in English)
Keyword(1) HMM-based speech synthesis
Keyword(2) rich context model
Keyword(3) parameter generation
Keyword(4) tree-based context clustering
1st Author's Name Shinnosuke TAKAMICHI
1st Author's Affiliation Nara Institute of Science and Technology()
2nd Author's Name Tomoki TODA
2nd Author's Affiliation Nara Institute of Science and Technology
3rd Author's Name Yoshinori SHIGA
3rd Author's Affiliation National Institute of Information and Communications Technology
4th Author's Name Sakriani SAKTI
4th Author's Affiliation Nara Institute of Science and Technology
5th Author's Name Graham NEUBIG
5th Author's Affiliation Nara Institute of Science and Technology
6th Author's Name Satoshi NAKAMURA
6th Author's Affiliation Nara Institute of Science and Technology
Date 2012-11-08
Paper # SP2012-78
Volume (vol) vol.112
Number (no) 281
Page pp.pp.-
#Pages 6
Date of Issue