Fast ICA for Multi-speaker Recognition System

It is hard to recognize speakers when the samples of voice are mixed. To overcome this shortcoming, this paper proposes a method: firstly, fast independent component analysis (Fast ICA) method is used for separating mixed voice signal of speakers. Secondly, a model of RBF neural network is used for speaker recognition. Based on the idea of blind signal separation, Fast ICA method can be used for signal separation because different voice signal source maintain a relatively independent identity. In this research of Multi-speaker recognition, the features can be extracted from the separated speech signals and a RBF neural network is used for the recognition model. Experiment results show that, this is an effective method for the mixed-voice speaker recognition.