From hybrid adjustable neuro-fuzzy systems to adaptive connectionist-basedsystems for phoneme and word recognition

This paper discusses the problem of adaptation in automatic speech recognit ion systems (ASRS) and suggests several strategies for adaptation in a modu lar architecture for speech recognition. The architecture allows for adapta tion at different levels of the recognition process, where modules can be a dapted individually based on their performance and the performance of the w hole system. Two realisations of this architecture are presented along with experimental results from small-scale experiments. The first realisation i s a hybrid system for speaker-independent phoneme-based spoken word recogni tion, consisting of neural net-works for recognising English phonemes and f uzzy systems for modelling acoustic and linguistic knowledge. This system i s adjustable by additional training of individual neural network modules an d tuning the fuzzy systems. The increased accuracy of the recognition throu gh appropriate adjustment is also discussed. The second realisation of the architecture is a connectionist system that uses fuzzy neural networks FuNN s to accommodate both a prior linguistic knowledge and data from a speech c orpus. A method for on-line adaptation of FuNNs is also presented. (C) 1999 Elsevier Science B.V. All rights reserved.