Voice conversion using the radial basis function network and the frequency–domain pitch modification technique

Abstract

This paper presents a novel algorithm for developing a voice conversion system that modifies the speech uttered by a speaker to sound as if produced by another target speaker. And an improved technique for a frequency–domain pitch modification is proposed. The approach is based on modifying spectral envelopes and the pitch of voiced sounds separately. The conversion rules for the spectral envelopes are constructed by the radial basis function (RBF) network, which is one of the well-known artificial neural networks. An excitation signal containing prosodic information is modified to match the average pitch using the proposed technique which proves that the speech quality deteriorates when the pitch is modulated by the Frequency–Domain Pitch-Synchronous Overlap and Add (FD-PSOLA) algorithm. The simulation results show that the proposed method achieves nearly optimal spectral conversion performance. Moreover, average cepstrum distance to the target speech is reduced by 87%, and the listening tests prove the proposed pitch modification technique maintains higher speech quality than the original FD-PSOLA.