Voice activity detection using multiresolution spectrum and support vector machines and audio mixing algorithm
Chengzhi Fang · 2006
Voice activity detection(VAD) uses Mel frequency cepstrum coefficient(MFCC) of multiresolution spectrum and two classical audio parameters as audio features,by which silence is prejudged by detection of multi-gate zero cross ratio,and audio features are classified by support vector machines(SVM).The real-time speech mixing algorithm uses short-time power of each audio stream as mixing weight vector.Experimental results show that the proposed VAD algorithm achieves overall better performance in all SNRs than the VAD of G.729B and other VADs.The output audio of the new speech mixing algorithm has excellent hearing perceptibility.Its computational time delay is short enough to satisfytheneedsofreal-timetransmission,andMCUcomputationislowerthanthatbasedonG.729BVAD.