A Case Study on Back-End Voice Activity Detection for Distributed Specch Recognition System Using Support Vector Machines

Recently, the Voice Activity Detection (VAD) algorithms based on machine learning techniques have shown impressive results in the area of speech recognition. In this paper, we present a case study and we discuss the performance of VAD based on Support Vector Machines (SVM) for Distributed Speech Recognition (DSR) system. In this case study, the speech and the non-speech frames are detected from the compressed Mel Frequency Cepstral Coefficients (MFCCs), at the back-end (e.g. Server) side, with the aim of improving the VAD performance and reducing the compression bit-rate from the front-end side. By using the trained SVM with polynomial kernel, the SVM-based VAD can produce encouraging detection results. The classification task conducted from the Aurora-2 speech database with different noise conditions shows comparable VAD performance, with respect to ETSI Advanced Front-End (ETSI-AFE) standard.