Imagine that you are in a room where two people are speaking at the same time. Two microphones are also placed in the same room, at different locations. Each of the microphones records a weighted sum of the speech signals transmitted by the speakers. It would be desirable to estimate the two original sources only from the microphone recordings. This is the problem we are trying to solve. It is commonly known as the Source Separation, and utilities a technique called Independent Component Analysis (ICA).

Independent Component Analysis was originally developed to separate speech signals, but due to the recent surge of research in the topic, it has been shown that ICA also has promising applications in telecommunications and medical signal processing.

In this project we will look at the present state of ICA and the techniques that are currently available. Algorithms such as Infomax, JADE and FastICA have proven to be very good estimates of sound signals, but are less successful at separating medical signals. Using MATLAB we will test the available algorithms on a variety of signals and analyse the strengths and weaknesses of each algorithm.

We hope to implement a functioning system, which can demonstrate source separation, showing its capabilities and limits.

David Sun, Dr Brian Krongold, Jessica Chen