Presentation | 2015-10-15 A study on quick model training in HMM-based speech synthesis Shuhei Yamada, Takashi Nose, Akinori Ito, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | In this paper, we propose an alternative model training technique using speaker-independent monophone models and decision trees to speed up the model training process of HMM-based speech synthesis. Recently, speech synthesis is used in various situations, and the demand of making speaker models by users themselves and synthesizing voices which reflects users' characteristic will increase. The HMM-based approach has an advantage of reflecting speaking styles and individualities to synthetic speech; however, it is not easy to train a reliable speaker-dependent model in a short time using prepared target speaker's speech because the model training process is computationally expensive compared with the synthesis process for it needs several time-consuming calculation such as decision tree construction. In the proposed technique, we use speaker-independent monophone models and decision trees instead of the conventional speaker-dependent ones. We also speed up the training process by skipping the parameter update based on maximum likelihood estimation. The objective and subjective experimental results show that the proposed training technique can synthesize speech having similar naturalness to the conventional training with less than 40% training time of the conventional one. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | HMM-based speech synthesi / model training / computation time reduction / speaker-independent models / context clustering |
Paper # | SP2015-64 |
Date of Issue | 2015-10-08 (SP) |
Conference Information | |
Committee | SP |
---|---|
Conference Date | 2015/10/15(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Kobe Univ. |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | Speech interface, Synthesis, Dialogue, Application system, etc. |
Chair | Kazunori Mano(Shibaura Inst. of Tech.) |
Vice Chair | Norihide Kitaoka(Tokushima Univ.) |
Secretary | Norihide Kitaoka(Tokyo City Univ.) |
Assistant | Takashi Nose(Tohoku Univ.) / Taichi Asami(NTT) |
Paper Information | |
Registration To | Technical Committee on Speech |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | A study on quick model training in HMM-based speech synthesis |
Sub Title (in English) | |
Keyword(1) | HMM-based speech synthesi |
Keyword(2) | model training |
Keyword(3) | computation time reduction |
Keyword(4) | speaker-independent models |
Keyword(5) | context clustering |
1st Author's Name | Shuhei Yamada |
1st Author's Affiliation | Tohoku University(Tohoku Univ.) |
2nd Author's Name | Takashi Nose |
2nd Author's Affiliation | Tohoku University(Tohoku Univ.) |
3rd Author's Name | Akinori Ito |
3rd Author's Affiliation | Tohoku University(Tohoku Univ.) |
Date | 2015-10-15 |
Paper # | SP2015-64 |
Volume (vol) | vol.115 |
Number (no) | SP-253 |
Page | pp.pp.27-32(SP), |
#Pages | 6 |
Date of Issue | 2015-10-08 (SP) |