Voice conversion through vector quantization.
作者： Masanobu Abe , Satoshi Nakamura , Kiyohiro Shikano , Hisao Kuwabara
DOI: 10.1250/AST.11.71
关键词: Speech recognition 、 Overall performance 、 Active listening 、 Vector quantization 、 Distortion 、 Computer science 、 Power (physics) 、 Pitch Frequency
摘要: A new voice conversion technique through vector quantization and spectrum mapping is proposed. This based on codebooks which represent the cor respondencebetween different speakers' codebooks. The for parameters, power values, pitch frequencies are separately generated using training utterances. makes it possible to precisely control individuality. performance of this confirmed by distortion frequency difference. To evaluate overall technique, listening tests carried out two kinds conversions: one between male female speakers, other speakers. In male-to-female experiment, all converted utterances judged as female, in male-to-male conversion, 57% them identified target speaker.
elsevier.com
sci-hub.se 参考文章(3)
Chikio Hayashi, RECENT THEORETICAL AND METHODOLOGICAL DEVELOPMENTS IN MULTIDIMENSIONAL SCALING AND ITS RELATED METHODS IN JAPAN Behaviormetrika. ,vol. 12, pp. 67- 79 ,(1985) , 10.2333/BHMK.12.18_67
D. Childers, B. Yegnanarayana, Ke Wu, Voice conversion: Factors responsible for quality international conference on acoustics, speech, and signal processing. ,vol. 10, pp. 748- 751 ,(1985) , 10.1109/ICASSP.1985.1168479
K. Shikano, Kai-Fu Lee, R. Reddy, Speaker adaptation through vector quantization international conference on acoustics, speech, and signal processing. ,vol. 11, pp. 2643- 2646 ,(1986) , 10.1109/ICASSP.1986.1168676