A novel SVM Kernel with GMM super-vector based on bhattacharyya distance clustering plus within class covariance normalization

A novel SVM Kernel based on Bhattacharyya distance clustering and within class covariance normalization was proposed to solve the problems of high computational complexity and susceptibility in channel interference of speaker verification. In our method, we computed the Bhattacharyya distance between pair of GMMs firstly. And then, a clustering algorithm was designed according to their Bhattacharyya distance to obtain clustering center models. MAP was applied on these clustering center models to generate super-vectors sequence kernel. Finally, within class covariance normalization was utilized to restrain the noise and channel distortion in this new kernel space. The experiment results showed that our proposed kernel has superior recognition accuracy and better robustness.