Voice conversion using dynamic inter-frame features

In conventional Gaussian mixture model(GMM)-based voice conversion systems,speech quality of converted utterances is degraded by over-smoothing of the predicted spectrum.A conversion method using dynamic inter-frame features was developed to alleviate the over-smoothing by taking account the continuity and variations of the object function.As a result,the predicted features are continuous and the variance is maximized into one syllable.Experimental results show that the method improves the opinion score of converted speech quality from 3.11 to 3.89,while effectively changing the speaker's individuality which shows that the dynamic features are important for quality voice conversion.