Voiceprint identification based on model clustering

A voiceprint identification method is proposed which be based on the speaker model clustering (SMC). The similar speaker models are clustered through an approximated KL divergence and determine the clustering center and their representatives to construct a hierarchical voiceprint identification model. During the recognition stage, a cluster is selected first by calculating distance between the test vector and clustering center or cluster representatives, then, though computing the logarithmic likelihood between the test vector and the speaker models in the selected cluster, the speaker can be determined with a significant decrease in the amount of computation. The experimental results show that the proposed method can improve the recognition speed about four times faster with a compromise of the accuracy rate as low as 0.95% compared with the traditional Gaussian Mixture Model (GMM). As a conclusion, the SMC method can improve the recognition speed with almost the same accuracy.