5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

A Novel Text-Independent Speaker Verification Method Using the Global Speaker Model

Yiying Zhang, Xiaoyan Zhu

Department of Computer Science, Tsinghua University, China

In this paper a new text-independent speaker verification method is proposed based on likelihood score normalization and the global speaker model, which is established to represent the universal features of speech and environment, and to normalize the likelihood score. As a result the equal error rates are decreased significantly, verification procedure is accelerated and system adaptability is improved. Two possible ways of establishing the global speaker model, one of which can meet the real-time requirement, are also suggested and discussed. Experiments demonstrate the effectiveness of this novel verification method and its improvement over the conventional method and other normalization methods.

Full Paper

Bibliographic reference.  Zhang, Yiying / Zhu, Xiaoyan (1998): "A novel text-independent speaker verification method using the global speaker model", In ICSLP-1998, paper 1144.