This item is available under a Creative Commons License for non-commercial use only
Electrical and electronic engineering
In recent years source separation has become an increasingly popular area of research in the signal processing community. The subject has found applications in a variety of fields such as medical imaging, sound and audio, econometrics and geophysics. This document will discuss the application of source separation techniques to the area of audio. Sound source separation is the process of observing a mixture signal made up of a number of sources, and from this mixture estimating the individual source signals. Audio source separation techniques may be crudely split into the following areas; techniques that utilise attributes of the sources, and that mimic methods used by the human auditory system to perform separation; and statistical, mathematical methods which do not necessarily take advantage of the attributes of sources. A further division is also possible whereby techniques utilise prior knowledge of sources, and those that do not, known as blind separation techniques. The novel work presented in this document discusses an approach for performing blind separation on a single channel mixture. The technique utilises attributes of the environment in which the the signal was recorded, and combined with the ADRess source separation algorithm, a novel process for source separation is presented.
Leddy, M.: Blind Single Channel Sound Source Separation. Masters Thesis. Dublin Institute of Technology, 2010.