Efficient Recognition of Speech Process using Non-deterministic Approach

Z. Ciota and G. Szymański (Poland)


Natural language processing, Fuzzy logic, Markov models, Neural networks


Design process of speech recognition has been presented. Going over the nowadays solutions of neural networks, fuzzy logic and Markov processes, several smart realizations of speech recognition have been discussed. The first one, how to recognize efficiently a whole word from the given vocabulary taking into account tree structures as a background of decision algorithm. Since the spoken language can be very often ambiguous, then it was useful to include fuzzy approach to obtain more effective methods of speech processing. In the most popular methods of speech processing Markov models are also applied. Hidden Markov model of order k estimates the probability of occurrence of a value in a given position basing on the sequence of k preceding values which are constituted in the learning process. The paper presents the application of hidden Markov models to text recognition in Polish language. It also includes the program description and results discussion.

Important Links:

Go Back