Probabilistic Integration of Joint Density Model and Speaker Model for Voice Conversion

摘要: This paper describes a novel approach to voice conversion using both joint density model and speaker model. In studies, approaches based on Gaussian Mixture Model (GMM) with probabilistic densities of vectors source target speakers are widely used estimate transformation. However, for sufficient quality, they require parallel corpus which contains plenty utterances the same linguistic content spoken by speakers. addition, GMM methods often suffer from over-training effects when amount training data is small. To compensate these problems, we propose integrate formulation. The proposed method trains few utterances, non-parallel target, independently. It eases burden speaker. Experiments demonstrate effectiveness method, especially Index Terms: conversion, model, unification