ERROR CORRECTION THROUGH LIKELIHOOD COMPENSATION USING PHONETIC BIGRAM FOR SPEAKER-INDEPENDENT ISOLATED WORD RECOGNITION

We propose an error correction technique for speakerindependent isolated word recognition by compensating for a word’s likelihood. Likelihood is compensated for by likelihood calculated by a phonetic bigram. The phonetic bigram is a phoneme model expressing frame correlation within an utterance. A speaker-independent isolated word recognition experiment showed that our proposed technique reduces recognition error compared to conventional techniques. The proposed technique achieves performance almost equal that without speaker adaptation compared to the conventional phoneme model adapted using several words.