Speech enhancement of noisy target voice for speaker-adaptive HMM-based speech synthesis

Abstract

In this paper, we propose a speech enhancement method for speaker-adaptive hidden Markov model (HMM)-based speech synthesis using noisy target voice. The proposed method is based on perceptually weighted multi-band spectral subtraction which can reduce musical noise in spectral components of estimated clean speech. It is shown from the experiment that an HMM-based speech synthesis system employing the proposed speech enhancement method can synthesize the target voice with good quality, compared to that directly using noisy target voice.