DocumentCode
442171
Title
Learning Mandarin tone mapping codebook for voice conversion
Author
Zuo, Guo-Yu ; Chen, Yao ; Ruan, Xiao-gang ; Liu, Wen-Ju
Author_Institution
Inst. of Artificial Intelligence & Robotics, Beijing Univ. of Technol., China
Volume
8
fYear
2005
fDate
18-21 Aug. 2005
Firstpage
4824
Abstract
The phonetics properties of Chinese Mandarin features clearly demarcated syllables with tonal phonemes. A tone codebook mapping method is proposed to obtain a better performance in voice conversion of Chinese Mandarin speech than the conventional conversion method, which focuses mainly on the short-time spectral envelopes. The pitch contour of the whole Mandarin syllable is used as a unit type for pitch conversion. The syllable pitch contours are first extracted from the source and target utterances. Time normalization and moving average filtering are then performed on them. These preprocessed pitch contours are classified to generate the source and target tone code-books, and by associating them a Mandarin tone mapping codebook is finally obtained in terms of speech alignment. Experimental results show that the proposed tone codebook mapping method in voice conversion can give a good performance in Chinese Mandarin speech.
Keywords
filtering theory; moving average processes; pattern classification; speech coding; Chinese Mandarin speech; moving average filtering; phonetics properties; short-time spectral envelopes; time normalization; tone codebook mapping method; voice conversion; Artificial intelligence; Filtering; Intelligent robots; Learning; Loudspeakers; Motion pictures; Radio broadcasting; Robotics and automation; Speech analysis; Speech synthesis; Codebook Mapping; Mandarin Tone; Voice Conversion;
fLanguage
English
Publisher
ieee
Conference_Titel
Machine Learning and Cybernetics, 2005. Proceedings of 2005 International Conference on
Conference_Location
Guangzhou, China
Print_ISBN
0-7803-9091-1
Type
conf
DOI
10.1109/ICMLC.2005.1527792
Filename
1527792
Link To Document