A Support Vector Machine Based Voice Activity Detection Algorithm for AMR-WB Speech Codec System

This paper proposed a new voice activity detection (VAD) algorithm using support vector machine (SVM) for improving the VAD performance of AMR-WB speech codec. The SVM is applied to train an optimized non-linear decision rule involving the VAD parameters, e.g., sub-band signal level, pitch gain, background noise level, and etc., defined in AMR-WB standard. Then, by the use of the trained SVM, the proposed algorithm can achieve accurate VAD under various noisy conditions. Experimental results carried out on the real speech signals show that the performance of the proposed VAD algorithm is better than that of AMR-WB VAD.