| Voice conversion technology converted the original speaker’s speech pattern into a target speaker’s speech pattern, keep the original semantic information unchanged in the process, and the converted voice is like what the target speaker said. Voice conversion is a new branch of the speech signal processing, the technology can be used in text-to-speech system, film dubbing, secure communication and so on, it has very important research value. This paper will mainly focus on converting speech spectral envelope conversion and prosodic feature conversion, and through the related problems are analyzed, get a complete system, implementation of voice conversion.This paper mainly focuses on several aspects as follows:(1) The relevant knowledge of voice conversion is studied, including the production of speech, mathematical model, commonly used in speech signal analysis method. This paper introduces the basic speech conversion system, and which used in the experiment of STRAIGHT model and the conversion performance evaluation criteria are discussed.(2) Through the analysis of commonly used spectral envelope conversion methods, this paper raises the selected spectral envelope conversion which based on mixed Gauss model, and the relevant problems of conversion and conversion steps are expounded.(3) Because of the neglect of super-segmental features research and conversion in the traditional voice conversion method, this paper focuses on the research of speech prosody conversion, put forward to carry out rich prosody conversion method. The prosodic features of conversion included fundamental frequency, speed, pause, stress.(4) This paper presents the overall framework of the system, and programming.Voice quality after conversion are evaluated from subjective and objective two respects, the experiment results show that the proposed speech conversion system converts the better performance than the conventional method. |
PDF Full Text Request