Third International Conference on Spoken Language Processing (ICSLP 94)
Automatic speech recognition ( ASR ) systems are made up of a number of different knowledge sources (KSs) which combine to solve the overall problem of speech recognition. An investigation into the benefits of each KS and the benefits of combining the KSs has been undertaken. Test sentences, taken from real data, have been processed by different versions of a speech recognition sub-system. The system performance was measured accurately by identifying the exact location of the required hypothesis (RH) in the hypothesis list (HL), at all cycles in the processing. This paper describes the analysis carried out and presents the results of this analysis and shows that each KS does play a role, though their importance varies considerably. The overall aim of this work is to aid in the development of a speech recognition system being produced at the University of Durham.
Bibliographic reference. Johnson, Kevin / Garigliano, Roberto / Collingham, Russell J. (1994): "Data-based control of the search space generated by multiple knowledge bases for speech recognition", In ICSLP-1994, 2147-2150.