Fuzzy-based clustering of speech recognition database

Algorithms have been proposed and experimentally tested for word recognition as part of a speech recognition system. These algorithms are based on a fuzzy model relying on the output signal of a recogniser of phoneme level. The recognition is carried out in a clustered database of words containing both the pronounced and the written forms of words as well. The clusters are selected on the basis of word schemes that characterise all of the words in a cluster and discriminate different clusters. Experiments have been made to prove the effectiveness of the proposed algorithms. The obtained improvement in speed was in our test approximately 30 times higher compared to a full search, without significant degradation of word recognition ratio