5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

A High-Performance Text-Independent Speaker Identification System Based on BCDM

Qin Jin, Luo Si, Qixiu Hu

Tsinghua University, China

This paper describes a Text-Independent Speaker Identification System of high performance. This system includes two subsystems, one is the close-set speaker identification system; the other is the open-set speaker identification system. In the implementation of the Text-Independent Speaker Identification System we introduce an advanced VQ method and a new distance estimation algorithm called BCDM (Based on Codes Distribution Method). In the close-set identification, the Correct Recognition Rate is 98.5% as there are 50 speakers in the training set. In the open-set identification, the Equal Error Rate is 5% as there are 40 speakers in the training set.

Full Paper

Bibliographic reference.  Jin, Qin / Si, Luo / Hu, Qixiu (1998): "A high-performance text-independent speaker identification system based on BCDM", In ICSLP-1998, paper 1112.