METHOD AND DEVICE FOR STRUCTURING SOUND MODEL, METHOD AND DEVICE FOR RECOGNIZING SPEECH, SOUND MODEL STRUCTURING PROGRAM, SPEECH RECOGNIZING PROGRAM, AND RECORDING MEDIUM WHERE SAME PROGRAMS ARE RECORDED

摘要

PROBLEM TO BE SOLVED: To improve speech recognition precision of a speech including reverberation. SOLUTION: A speech signal observed in an environment different from an object of speech recognition is subjected to reverberation removal and a sound model parameter of a speech signal having its reverberation removed is estimated to structure a sound model according to the estimation result of the sound model parameter. In speech recognition, reverberation is removed from a speech to be recognized, a sound model is adapted with the speech having the reverberation removed, and distortion caused in the reverberation removal is removed to make speech recognition, thereby the recognition precision is improved. COPYRIGHT: (C)2006,JPO&NCIPI