Accent extraction of emotional speech based on modified ensemble empirical mode decomposition

Abstract

Ensemble empirical mode decomposition (EEMD) is a noise-assisted adaptive data analysis method to solve the mode mixing problem caused by empirical mode decomposition (EMD), which is a significant step of Hilbert-Huang Transform (HHT). In this paper, a novel fast EEMD preferences algorithm called Quasi-Gradient Search (QGS) is proposed. For a given ensemble number, we first apply Nonlinear Correlation Coefficient (NCC) to estimate the lower bound of decomposition error, which leads to the best amplitude of added noise. According to the accuracy requirement, we can obtain the minimum ensemble number to solve mode mixing by increasing the ensemble number exponentially. Furthermore, the QGS is applied to extract the accents of the emotion speeches in different scales to solve the mode mixing problem. Compared with the result of traditional EEMD, the proposed QGS can greatly enhance the calculation speed with the same decomposition accuracy.