ACOUSTIC MODEL RETRAINING METHOD AND APPARATUS USING MULTI-LANGUAGE JOINT PHONE SET

摘要

The present invention relates to a retraining method by complementing insufficient audio data for a particular language and without changing the structure of the acoustic model simultaneously using a joint phone which is included in the multilingual audio data. Speech data for each language must be fully prepared in order to create an acoustic model of a multi-continuous speech recognition device but consumes a lot of costs and time in general. And the present invention can define common phonemes to be used for a phoneme symbol which is acoustically same between multiple languages. Provided is a selective retraining method using the common phoneme in the language which has a large amount of the audio data in order to train the language of insufficient memory on the basis of above.