JPS6470800A - Voice recognition equipment - Google Patents

Abstract

PURPOSE: To reduce detection errors between sounding sections and to recognize voice at a high recognition rate by providing a proper restriction between endpoint free sections. CONSTITUTION: A part 1 for detecting voice sections detects such a voice section as the amplitude level of the input wave S is larger than a prescribed threshold, while the detection signal K is 1 within the sections. A part 2 for detecting stable characteristic points detects characteristic points in the sections where the signal K is 1; making Ax a starting end candidate section, Ax being the front and the rear of the point moved against from the first characteristic point A2 of an input pattern (a) by the length L1 between the starting end B1 and the first characteristic point B2 of the standard pattern (b); making Ar a completing end candidate section, Ar being the point before the point advanced by the length L2 between the last characteristic point B3 and the completing end B4 of an input pattern (b); recognizing the voice of the input wave S with the starting end/completing end free, and outputting the result R. With the characteristic points A2 and B2 coincided with each other on a time base, and with A3 and B4 coincided while the pattern (b) is expanded and contracted, a recognizing part 3 recognizes voice based on the recognition principle. With this structure, voice can be stably recognized at a high recognition rate even in the presence of unstationary noise.