A KERNEL METHOD FOR SPEAKER RECOGNITION WITH LITTLE DATA

It proposed a fuzzy kernel vector quantization method for speaker recognition with little training data. By non-linear mapping, it quantized the input data in the high-dimensional feature space, and used the cluster centers to form the speaker´s model. Because of the kernel method, it made the inherent speech features explored, and the dissimilarity among different speakers increased. Besides, it used the fuzzy kernel nearest prototype classifier to identify unknown speech. Experimental results show that the performance of this method is better than fuzzy vector quantization and Gaussian mixture model method when training data is little or limited