Frequency warping based on mapping formant parameters

Abstract

We propose a novel method of generating a frequency warping function by mapping formant parameters of the source speaker and the target speaker. Alignment and selection process are performed to ensure that the mapping formants can represent speakers' difference well. This approach requires only a very small amount of training data for generating the warping function, which can greatly facilitate its application. It can also achieve high quality of the converted speech while successfully converting a speaker's identity. A practical voice morphing system has been built based on this approach. And experimental results show its effectiveness.