Voice morphing using the generative topographic mapping

In this paper we address the problem of Voice Morphing. We attempt to transform the spectral characteristics of a source speaker's speech signal so that the listener would believe that the speech was uttered by a target speaker. The voice morphing system transforms the spectral envelope as represented by a Linear Prediction model. The transformation is achieved by codebook mapping using the Generative Topographic Mapping, a non-linear, latent variable, parametrically constrained, Gaussian Mixture Model.