HMM-based Tibetan Lhasa speech synthesis system

Author

Wu Zhiqiang ; Yu Hongzhi ; Li Guanyu ; Wan Shuhui

Author_Institution

Key Lab. of China´s Nat. Linguistic Inf. Technol., Northwest Univ. for Nat., Lanzhou, China

fYear

2013

fDate

12-13 Oct. 2013

Firstpage

92

Lastpage

95

Abstract

According to the pronunciation characteristics of Tibetan Lhasa dialect, this design realizes the Tibetan Hidden Markov Model(HMM) parametric speech synthesis system. At first, we prepare data from several aspects, such as text design and recording, speech segmentation, tagging, problem sets. Then we built the Tibetan Lhasa dialect pronunciation training model and synthetic speech sound by using MLSA speech synthesizer. Finally we evaluate the quality of synthetic by using the methods of mean opinion score (MOS), the evaluation results show that the mean opinion score (MOS) was around 3.2 which basically achieve the anticipated design effect.

Keywords

hidden Markov models; natural language processing; speech synthesis; HMM-based Tibetan Lhasa speech synthesis system; MLSA speech synthesizer; MOS; Tibetan Lhasa dialect; Tibetan Lhasa dialect pronunciation training model; Tibetan hidden Markov model parametric speech synthesis system; methods mean opinion score; problem sets; pronunciation characteristics; speech segmentation; synthetic speech sound; tagging; text design; text recording; Educational institutions; Hidden Markov models; Rhythm; Speech; Speech synthesis; Training; Tibetan; hidden Markov model; speech synthesis; synthesis by parametric modeling;

fLanguage

English

Publisher

ieee

Conference_Titel

Computer Science and Network Technology (ICCSNT), 2013 3rd International Conference on

Conference_Location

Dalian

Type

conf

DOI

10.1109/ICCSNT.2013.6967070

Filename

6967070