Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

Dynamic Machine Adaptation in a Multi-Speaker Isolated Word Recognition System

Sekharjit Datta

Department of Electronic & Electrical Engineering, Louhghborough University of Technology, Loughborough, UK

This paper investigates a dynamic adaptation technique applied to a real-time isolated word recognition system which continually adapts to the speaker's characteristics and improves the performance of the system. The technique initially allows the system to compensate for inadequate training of the classifier and then continues to track changes in the speaker's voice. The overall results show that a recognition rate of over 96% is achievable when tested on 12 unknown speakers. The study also reveals that the crucial factors in a dynamically adaptive speech recognition are: (a) knowing when to adapt and (b) deciding by how much to adapt.

Full Paper

Bibliographic reference.  Datta, Sekharjit (1994): "Dynamic machine adaptation in a multi-speaker isolated word recognition system", In ICSLP-1994, 2223-2226.