Excitation Modeling Method Based on Inverse Filtering for HMM-Based Speech Synthesis

In this paper, we propose a novel excitation modeling approach for HMM-based speech synthesis system (HTS). Here, the excitation signal obtained via inverse filtering is parameterized into excitation features, which are modeled using HMMs. During synthesis, the excitation signal is reconstructed by modifying the natural residual segments in accordance with the target source features generated from HMMs. The proposed approach is incorporated into the HTS. Subjective evaluation results indicate that the proposed method enhances the quality of synthesis and is better than the traditional pulse and STRAIGHT-based excitation models.