Sixth European Conference on Speech Communication and Technology

Budapest, Hungary
September 5-9, 1999

A Neural Network-Based Text-Dependent Speaker Verification System Using Suprasegmental Features

M. Mathew, B. Yegnanarayana, R. Sundar

Department of Computer Science and Engineering, Indian Institute of Technology, Madras, India

In this paper, we propose two neural network-based approaches, namely, One Speaker One Network and One Speaker Multiple Networks, for text-dependent speaker verification using suprasegmental features. The suprasegmental features used for this study are pitch accent and durational features. These features are extracted using properties of intonation patterns and duration. We have proposed an approach to combine evidence present at the segmental and suprasegmental levels to improve the performance of the verification system.

Full Paper (PDF)   Gnu-Zipped Postscript

Bibliographic reference.  Mathew, M. / Yegnanarayana, B. / Sundar, R. (1999): "A neural network-based text-dependent speaker verification system using suprasegmental features", In EUROSPEECH'99, 995-998.