Voice activity detection in car environment using support vector machine and wavelet transform

In this paper, a new voice activity detection (VAD) algorithm using wavelet and support vector machine (SVM) is developed for improving the performance of speech coding or recognition systems working in car environment. The first step of the proposed algorithm is to apply wavelet-based de-noising method to reduce the in-car background noise of the input speech. Then one can make use of SVM to train an optimized non-linear decision rule involving the subband power, zero-crossing rate, and pitch frequency of the de-noised input speech. By the use of the trained SVM, the proposed algorithm can achieve accurate VAD under in-car noisy environment. Various experimental results carried out on the Aurora speech database show that the proposed SVM-VAD is capable of outperforming to the standard VADs including ITU G.729B and ETSI AMR-NB VAD.