Impact of emotional speech to automatic speaker recognition - experiments on gees speech database

Abstract

© Springer International Publishing Switzerland 2014. In this paper an automatic speaker recognizer is tested on a speech database with emotional speech. The automatic speaker recognizer is based on mel-frequency cepstral coefficients as features of a speaker and covariance matrices as speaker models. The speaker models are trained on one sentence of neutral speech for each speaker. Other sentences from the same speech database are used for testing, including both neutral and four emotional states: happiness, fear, sadness, and anger. The aim of research is to investigate the impact of acted emotional speech to accuracy of automatic speaker recognition.