Investigation impact of environment for performance of ICA for speech separation

Abstract

Speech signals extraction is becoming more and more important in variety applications such as mobile phones, conference equipments, radars and others. This paper presents a blind method to enhance and separate speech signals in noisy environment. The proposed technique exploits the independent component analysis (ICA) to separate speech signals. We investigate impact of environment (noise, number of sources and their location, etc.) for performance of this method. For evaluation of performance are used parameters source-to-interference ratio SIR, source-to-distortion ration SDR and source-to-artifacts ration SAR.