Comparisons of speaker recognition strengths using suprasegmental duration and intensity variability: an artificial neural networks approach

Abstract

This study compares the speaker recognition strengths based on suprasegmental duration and intensity variability in the speech signal using artificial neural networks. Such algorithm can well capture the nonlinear effects in the data, and is more robust against noise in the data. Three rounds of classification tasks were performed with 1) duration metrics, 2) intensity metrics, and 3) the combination of duration and intensity metrics as the independent variables. The results indicated that both intensity and combined metrics significantly outperformed the duration metrics. Moreover, the combination of intensity and duration metrics showed higher probability of improved speaker classifications than intensity metrics over duration metrics.