Cepstrum liftering based voice conversion using RBF and GMM

Voice Conversion is a technique which morphs the speaker dependent acoustical cues of the source speaker to those of the target speaker. Speaker dependent acoustical cues are characterized at different levels such as shape of vocal tract, glottal excitation and long term prosodic parameters. In this paper, low time and high time liftering is applied to the cepstrum to separate the vocal tract and glottal excitation of the speech signal. The Radial Basis Function and Gaussian Mixture Model are developed to capture the mapping functions for modifying the cepstrum based vocal tract and glottal excitation of the source speaker according to a target speaker. The subjective and Objective measures are used to evaluate the comparative performance of RBF and GMM based voice conversion system. Results indicate that the RBF based transformation can be used as an alternative to GMM based model. Subjective evaluation illustrate that the proposed algorithm maintains target voice individuality, quality of the speech signal.