First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

Speaker Recognition Using Static and Dynamic CEPSTRAL Feature by a Learning Neural Network

Hujun Yin (1), Tong Zhou (2)

(1) Dept. of Electrical Engineering, Tongji University, Shanghai, P.R.China

(2) Computer Laboritory, Shanghai TV University, Shanghai, P.R.China

We have applied a Multi-layer learning neural network to speaker recognition. A set of LPC-based cepstral coefficients and their orthogonal polynomials were chosen as static and dynamic spectral feature of speaker's utterance. A number of experiments have been conducted to assess the performance of the network of both text-dependent and text-independent speaker recognition tasks. The results describe the relations between the recognition accuracy and the number of hidden units, the number of training set presentations, and the number of speaker database. In a critical condition, the network achieved a high recognition rate of 98.5 percent correct and 95.2 percent correct in text-dependent and text-independent tests respectively.

Full Paper

Bibliographic reference.  Yin, Hujun / Zhou, Tong (1990): "Speaker recognition using static and dynamic CEPSTRAL feature by a learning neural network", In ICSLP-1990, 1277-1280.