Excitation Modeling for HMM-based Speech Synthesis Based on Principal Component Analysis

This paper proposes a new excitation modeling method for improving the quality of HMM-based speech synthesis. The proposed excitation or source modeling method models the pitch-synchronous residual frames extracted from the excitation signal. Initially, principal component analysis is performed on the pitch- synchronous residual frames. Based on the analysis, the pitch synchronous residual frames are parameterized in two stages. In first stage, the dominant component of the residual frame is represented using PCA coefficients and in the second stage, the noise component of the residual frame is parameterized in terms of spectral and amplitude envelopes. The proposed excitation model is integrated into HMM-based speech synthesis system. Subjective evaluation results indicate that the speech synthesized by the proposed excitation model is significantly better than the two existing excitation modeling methods.