System And Method of Pronunciation Variation Modeling Based on Indirect data-driven method for Foreign Speech Recognition

摘要

A modeling system and a method thereof for ratio direct data base utterance side for improving the performance of the voice recognition system are provided to extract ratio direct data base utterance and improve the performance of the voice recognition system. A speech recognition modeling system comprises a characteristic detecting unit(101), a sound model database, an utterance model database, and a voice recognition circuit(102). The feature vector from the inputted voice is extracted from the characteristic detecting unit. The audio data are stored the sound model database. Pronunciation data are stored the utterance model database. The voice recognition circuit finds the word string in which the phoneme probability is the highest from the word string based on the feature vector.