METHOD AND APPARATUS FOR SPEECH REGISTRATION

摘要

A method and an apparatus for recognizing voice are provided to register and recognize new words in real time through two way voice recognition based on language model correction, thereby increasing accuracy of voice recognition. A feature vector extractor(110) divides an inputted voice signal(10) into a unit frame to extract a feature vector corresponding to a divided frame area. The first phoneme recognition unit(120) produces a phonetic lattice from the feature vector of the inputted voice signal. The first phoneme recognition unit produces the first phoneme row group by recognizing a phoneme by pattern-matching the feature vector by the first sound model and language model. The first phoneme recognition unit produces a phonetic lattice from the first phoneme row group. The second phoneme recognition unit(140) recognizes a phoneme by pattern-matching the feature vector by the second sound model and corrected phonetic lattice.