Human Voice Conversion Based On Parameter Models

In this paper, a voice conversion algorithm based on the parameter models is put forward, which can implement voice conversion through modifying the parameters of voice source and vocal tract model. On the basis of LF-4derivative glottal flow, an improved algorithm is presented to manipulate the voice source parameters flexibly. It also proposes an automatic parameterization method of the voice source, and synthesizes the voice source signal with these parameters. At the same time, the paper has discussed the transformation rule of the characteristic parameters of vocal tract following the differences of gender and physical status, and applied the frequency warping algorithm, which uses bilinear transform function to carry through the non-linear spectrum shifting, to converse the vocal tract model based on the vocal tract unit impulse response.Voice conversion between male speech and female speech is realized by appropriate control over the parameters of the voice source and the vocal tract model. The result and analysis of the experiment show that, this algorithm can synthesize speech with high naturalness, and the speech conversed from female to male is better than that from male to female.