On the performance of cepstral features for voice-based gender recognition

Abstract

Voice-based gender detection is an interesting problem. This article shows our attempts to observe the impact of various short-term spectral features with varying number of dimensions on gender recognition systems. We demonstrate our experiments on SITW and ELSDSR databases to determine the best combination of features for improved performance. An attempt has been made to investigate the effect of these systems under mismatched conditions, and it is seen that the existing scenario needs better algorithms to improve cross-corpus performance.