Wavelet based analysis of speech under stress

Abstract

Stress and its effects in speech signals have been studied by many researchers. A number of studies have been attempted to determine reliable acoustic indicators of stress using such speech production features as fundamental frequency (F0), intensity, spectral tilt, the distribution of spectral energy and others. The findings indicate that more work is necessary to propose a general solution. The goal of this study is to propose a new set of speech features based on dyadic wavelet transform (DyWT) coefficients as potential stress sensitive parameters. The parameters' ability to capture different stress types is evaluated on the basis of separability distance measure between parameters of each stress class for four stress conditions: neutral, loud, question and angry. After an extensive number of scatter distributions were considered a number of clear trends emerged that confirmed that the new speech parameters are well suited for stress classification.