Experiments of GMM based speaker identification

摘要

In human-robot interaction areas, the robot is often expected to recognize the identity of the speaker in some specific scenarios. It is a kind of biometric modality, and in general using statistical model is a classical and powerful method dealing with speaker identification problem. In this paper, we apply the Gaussian mixture model (GMM) on the speech feature distribution modeling and build the speaker identification system under MATLAB platform. Experiments are conducted on practical speech database and we also further give some insights into feature extraction, different length input utterances analysis and the impostor situation.