SIMILARITY METHOD OF DIVISION BASE FOR LOW COMPLEXITY SPEECH RECOGNIZER

摘要

PROBLEM TO BE SOLVED: To express a given speech utterance as a digital word model by expressing a word model with a vector of a value of the sum closely coordinating to the normalized similarity data.;SOLUTION: A speech normalization module 18 divides an input signal to a speech part and a non-speech part for normalizing the phoneme similarity data. Then, through the speech part is divided to the prescribed number of segments, the sum of all normalized phoneme similarity in one division in made to become equal at every segment. Then, a word modeling module 20 constitutes one word model from the normalized phoneme similarity data. Thus, in respective segment, by summing up all normalized phoneme similarity values answering to a specified phoneme, one value of the sum is decided. In such a manner, the word model is expressed by one vector of the sum closely coordinating to the normalized phoneme similarity.;COPYRIGHT: (C)2000,JPO