An real-time voice activity robust detection based on subband spectrum

摘要

This paper presents a complex environment to adapt to the efficient real-time endpoint detection algorithm, each frame is given in the acoustic signal in the noise power spectrum filtering the projection method. First spectrum of each frame iteration voice filtering, then it is divided into several sub-bands and calculate the entropy of each subband spectrum, and then have a number of sub-band spectral entropy of the frame after a median filter to obtain a set of Spectral entropy of each frame, according to the value of spectral entropy to classify the input voice. Experimental results show that the algorithm can distinguish speech and noise, can significantly improve the performance of speech recognition systems, in different environmental conditions, noise robust. The algorithm to calculate the cost of a small, simple and easy to implement for real-time voice recognition system.