HOME    About this site    mypage    Japanese    library    university    Feedback

University of the Ryukyus Repository >
Faculty of Engineering >
Doctoral Dissertation (Graduate School of Engineering and Science) >

 
Title :Myanmar Text-To-Speech System with Rule-based Tone Synthesis
Title alternative :ミャンマー語の声調の規則合成とテキスト音声変換システム
Authors :Win, Kyawt Yin
Issue Date :Mar-2011
Abstract :学位の種類(博士(工学)),学位授与機関(琉球大学),授与年月日(平成23年3月23日),学位記番号(理工研第208号)
The MyanmarTTS system is a rule-based speech synthesis system, in which fundamental speech units are demi-syllables with Level tone. We use Myanmar (Burmese) official language for the TTS system. To construct the TTS system, monosyllabic words are analyzed and the parameters are obtained for synthesis of Myanmar tones. This TTS system uses a source filter model and furthermore a Log Magnitude Approximation Filter is employed for the vocal tract character. Myanmar is a tonal language comprising 4 different tones. The Level and High falling tones belong to the long-tone group and the Falling and Checked tones belong to the short-tone group. Checked tone is accompanied by a glottal stop. In this study, normalization method is different from other TTS systems such as Thai and Vietnamese The synthesized tones are evaluated by employing listening tests. These results show that our rule-based linear pattern gives high intelligibility for the synthetic sound and is considered to be sufficient for MyanmarTTS system. The high intelligibility of synthesized tone was confIrmed through listening tests of synthesized words with correct rates 95.6% for male and 97.8% for female speech. As a result, we showed that linear pattern is sufficient for Myanmar tone synthesis. The introduced normalization method is applicable for other tone synthesis rule of other tonal languages. Large variability exists in the F_0 and the length uttered by deferent speakers and different syllables. Hence for speech synthesis normalization of F_0 and lengths are important to discriminate tones. Here, we implement tone rule by using two parameters; optimized F_0 and length. As an advantage in the proposed method, the optimized parameters can be separated to male and female group. The effectiveness of the proposed method is confirmed by the distribution of F_0 and length. Listening tests with high correct rates approve intelligibility of synthetic sound. Furthermore, in the proposed method, the optimized parameters can be separated into male and female groups. The introduced proposed method is applicable for other tone synthesis rule of other tonal languages.
Type Local :学位論文
Publisher :Win Kyawt Yin
URI :http://hdl.handle.net/20.500.12000/20526
Appears in Collections:Doctoral Dissertation (Graduate School of Engineering and Science)

Files in This Item:

File Description SizeFormat
riken208 .pdf1749KbAdobe PDFView/Open