A Fast Speaker Verification Method With Global Speaker Model

摘要

In this paper a fast text-independent speaker verification method is proposed based on lielihood score normalization. A global speaker model is established to represent the universal features of speech. Likelihood score is subtracted by the score produced by this model so that the common factors of speech are removed and the differences between speakers are brought into prominence. Compared to the conventional method and popular likelihood score normalization approach, the proposed technique decreases theequal error rate significantly, and the verification speed is fast.