The Research Of Voice Activity Detection In Engineering Machinery Noise Environment

Recent years，VOIP technology is booming,and used more widely.But now VOIPis mainly used in PC field,In embedded field is not yet used on a large scale.The mainreason is the situation is more complex in embedded field,Such as soundquality,network condition etc.In construction machinery field,the environment isworse than normal.human’s voice often coverd by mechanical equipment,and lead tocalls can’t be completed.To solve this problem,I focus researched the VAD(Voiceactivity detetion) algorithm.In the construction machinery strong noise environment, there have lots of noisesource, speech often covered by the machine’s noise, calls often can’t success andwaste bandwidth. To solve this problem, a new vad algorithm based on HMM andSVM(HMM/SVM-VAD)was proposed in this paper.HMM(Hidden Markov Model)cansimulate the speech model better,but perform generally in classificationfield.SVM(Support Vector Machine) perform better in classification field,butinconvenient to process the speech data.This method complement each other’sadvantages and disadvantages of the two.This method first input the MFCC into theHMM,Second,get the N-best Recognition results by using Viterbi algorithm,and thentransform the N-best recognition results to SVM feature vector. Last, use the SVM toget the classification results. Experiments demonstrate HMM/SVM-VAD can detectthe voice and noise very well in the machinery strong noise environment.