Enhancement of hearing-impaired Mandarin speech

This paper presents a new voice conversion system that modifies misarticulations and prosodic deviations of the hearingimpaired Mandarin speech. The basic strategy is the detection and exploitation of characteristic features that distinguish the impaired speech from the normal speech at segmental and prosodic levels. For spectral conversion, cepstral coefficients were characterized under the form of a Gaussian mixture model with parameters converted using a mapping function that minimizes the spectral distortion between the impaired and normal speech. We also proposed a VQ-based approach to prosodic conversion that involves modifying the features extracted from the pitch contour by orthogonal polynomial transform. Experimental results indicate that the proposed system appears useful in enhancing the hearing-impaired Mandarin speech.