Multilayer Knowledge Base System for Speaker-Independent Recognition of Isolated Words

Abstract

The aim of the presented research was to elaborate and test the speaker-independent system for the man-machine voice interfacing using small vocabulary (digits). Speaker-independent recognition of digits may be applied practically, among others in telecommunication for the dialing by voice. The mentioned assumptions caused the necessity to consider technologically applicable and economically optimized concepts only. Thus, extensive search for the theoretical background and relevant technologies was performed at the initial stage of the investigations. Resulting speech recognition system implements the combined methods of time-domain signal feature extraction, of neural network learning procedures and of fuzzy logic decision rules. All mentioned procedures tested at the software model stage are fully supported by the new hardware solutions in the domain of signal processing. More detailed analysis of the elaborated system reveals its multi-layer nature, hence it applies the domain-dependent knowledge base to acquire and to reorder data in such a way that they become separable and it uses learning algorithms to acquire knowledge enabling final object classifications.