Blind Source Separation Approach for Audio Signals based on Support Vector Machine Classification

摘要

Audio signals are surrounding us everywhere, existing in many forms (speech, music, noise background, ...), but they exist all mixed together and separating them is a real serious problem. It is required to arrange them in order to be separated to use them an easy way in such many various applications such as blind source separation, extraction of speech segments, audio visual analysis,.... In this work, we introduce a new method to separate audio signals arrived mixed to a couple of microphones implemented on a head of a humanoid robot to solve the blind source separation (BSS) problem using the support vector machine (SVM). Thus, we provide a theoretical introduction to present the SVM method which has frequently been proposed for classification and regression tasks. The observations are classified by SVM method using some standard recordings which have been taken in a room.