Development of statistically consistent voice conversion techniques based on joint feature modeling

2014 Fiscal Year Final Research Report

• Project/Area Number: 24700166
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Perception information processing/Intelligent robotics
• Research Institution: Nagoya Institute of Technology
• Principal Investigator: NANKAKU Yoshihiko 名古屋工業大学, 工学(系)研究科(研究院), 准教授 (80397497)
• Project Period (FY): 2012-04-01 – 2015-03-31
• Keywords: 声質変換

Outline of Final Research Achievements

This project aimed to improve voice cconversion techniques which convert speech waveforms from original speaker's voice to another speaker's one. In conventional voice conversion technique, spectral features and prosodic features such as fundamental frequencies (F0) and speaking rates are indenedently converted. In the proposed technique, these features are consistently modeled using a single statistical model and all features are jointly converted using the correlation among features. Experimental results showed that the speech quality of converted voices was improved by the proposed technique. Moreover, the project also developed a technique to improve voice conversion with a very small amount of training data is available.

Free Research Field

音声情報処理