METHOD AND APPARATUS FOR RECOGNIZING WORD SEQUENCE OF AUDIO SIGNAL

摘要

PURPOSE: To obtain a sound signal word sequence recognizing method to be more quickly driven without remarkably lossing the reliability of recognition by generating scoring to be individually stored together with a reference for a completion word allocated from a prescribed score before starting it at a bottom. CONSTITUTION: Continuous sound signals obtained from an acoustic sound signal through a microphone are scanned and timewisely individually supplied to a preprocessor 20 as a sequence of values 18 with time symbols i-2, i-1, i, i+1 and the preprocessor 20 determines a sequence of test signals respectively characterizing the sections of the sound signals. A unit 22 compares these test signals with a reference signal stored in a relative memory 24 and supplies the compared results to the output side as scores. These scores are used in a block 26 adaptively to dynamic time, a new score is supplied to a block 28 and compared with a threshold so that a partial hypothesis in which the score is more than the threshold is not continued.