A Voice Conversion System Based on the Harmonic plus Noise Excitation and Gaussian Mixture Model

This paper proposes an algorithm to extract the excitation source based on the Harmonic plus Noise Model, which decomposes the source voice to harmonic components and random noise component. By The proposed algorithm, the LPC parameters of the harmonic components are extracted through linear prediction method, and then the LPC inverse filter is used to get the harmonic residual signal as the excitation source. This excitation source avoids artificial modifications and contains more speaker personality characteristic. Finally the synthesized speech is superimposed on the random noise component compensation. Experiments demonstrate that proposed algorithm improves the target tendentiousness and naturalness of the synthesized speech.