PURPOSE: To make decision on sound presence/sound absence possible with high accuracy by reducing the influence of variation in input speech energy as in-use environment changes.
CONSTITUTION: A frame energy calculating circuit 32 divides a speech sending signal into frames and finds input energy SE(k), frame by frame, and a speech detecting circuit 34 calculates a speech frame metric SFM(k) and a noise frame metric NFM(k) respectively on the basis of the input energy SE(k), generates an adaption threshold value TM(k) which varies with the input energy SE(k), and generates a decision threshold value on the basis of the adaption threshold value TM(k) and noise frame metric NFM(k). Then, sound presence/sound absence decision is made, frame by frame, by comparing the levels of the decision threshold value and speech frame metric SFM(k) with each other.
WO/2018/013343 | AUDIO SLICER |
JPH02101500 | VOICE RECOGNIZING DEVICE |
OKUDA YUJI