Text and speaker independent voice conversion

Abstract

This paper describes an approach to the challenging problem of text and speaker independent voice conversion. The approach is based on target speaker’s speech production process parameterization using harmonic analysis. Unified model allows processing of any input speech regardless of its content and source speaker. The method provides subjective quality of conversion that is comparable with text/speaker depended methods though requires a drastically simplified teaching routine and a small codebook. The experimental results are given in the paper.