Research on Speech Feature Extraction and Synthesis Algorithm Based on EEMD

Abstract

In the speech synthesis system based on statistical acoustic modeling, the parameter generation method based on Hidden Markov Model (HMM) has been fully developed. The parameters are mainly Mel Frequency Cepstrum Coefficient (MFCC), so the accuracy of extracting the speech signal MFCC directly affects the accuracy of speech synthesis. In this paper, based on the traditional method of extracting MFCC features, the Ensemble Empirical Mode Decomposition (EEMD) algorithm is added, and the Mel filter bank in the traditional algorithm is changed to an adaptive Mel filter bank. The EEMD algorithm helps to enhance the ability of MFCC to describe the original speech signal, and the adaptive Mel filter bank can improve the utilization of the filter, thereby further improving the accuracy of MFCC. The experimental results show that compared with the traditional algorithm, the proposed algorithm can significantly reduce the MFCC root mean square error (RMSE) of the synthesized speech sequence.