Presentation 2012-11-08
Prosody Generation based on HMM using Tow-stage Clustering
Yasuyuki MITSUI, Reishi KONDO, Masanori KATO,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) HMM-based speech synthesis can generate highly natural prosody, but there is a problem that the pitch patterns represent the accent different from the designation might be generated. In this paper, for the purpose of reducing the error of accent caused by abnormality of pitch pattern outlines in prosody generated by HMM, we propose the learning method of prosody models introduced two-stage decision tree clustering. The set of questions to be used in the first stage of clustering is configuring by the only questions about pitch pattern outlines. Then, the tree structure near the root node is constructed with only the nodes that are split by the question about pitch pattern outlines. As a result of evaluation experiments, we confirmed that the proposed method decrease the error of accent in prosody generation by HMM by half with the decision tree has the same sized structure of the conventional method.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Speech Synthesis / HMM / Prosody Generation / Decision Tree / Context Clustering
Paper # SP2012-80
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) Prosody Generation based on HMM using Tow-stage Clustering
Sub Title (in English)
Keyword(1) Speech Synthesis
Keyword(2) HMM
Keyword(3) Prosody Generation
Keyword(4) Decision Tree
Keyword(5) Context Clustering
1st Author's Name Yasuyuki MITSUI
1st Author's Affiliation Information and Media Processing Labs., NEC Corporation()
2nd Author's Name Reishi KONDO
2nd Author's Affiliation Information and Media Processing Labs., NEC Corporation
3rd Author's Name Masanori KATO
3rd Author's Affiliation Information and Media Processing Labs., NEC Corporation
Date 2012-11-08
Paper # SP2012-80
Volume (vol) vol.112
Number (no) 281
Page pp.pp.-
#Pages 6
Date of Issue