SYSTEM CLUSTERING METHOD OF VOICE RECOGNITION FOR THE ARBITRARY WORD

摘要

The present invention relates to a clustering method of a speech recognition system for arbitrary word recognition in which an optimal question is selected at each node so that distortion can be reduced as a whole.;The clustering method of the speech recognition system for any word recognition of the present invention comprises the steps of setting and storing the entire trained speech parameter as an element of the root node; Extracting one node to select an optimal question for the node from a question set, separating the plurality of nodes into a plurality of nodes, and storing the selected questions; And repeating the above two steps until a certain stop criterion is satisfied, and four steps of determining the stored node as the final leafset if the random stop criterion is satisfied.;According to the present invention, by clustering in consideration of both the distortion at the current node as well as the distortion with other clusters, there is an effect that not only the distortion is reduced as a whole, but also the discrimination between the clusters as a whole.