Audio Zoom: blind source separation in a reverberant environment

Dr. Peter Driessen’s research group at University of Victoria is working on building a system using microphone arrays of arbitrary geometry to zoom in on desired audio in a noisy environment. Ultimately a listener will be able to simply look at the location from where s/he wants to hear the audio, controlling the audio zoom via head and eye gestures.

Applications of this new audio zoom system include a super hearing aid for people in a crowded noisy environment. Such a hearing aid will have performance far exceeding any standard hearing aid with microphones near the ears. Audio zoom may be very useful for the film industry, to capture better quality audio during on-location filming, and reduce the amount of re-recording and post-production required. It may also be useful for the computer games industry where zooming on natural sounds may be desired as part of the game play. Audio zoom will be a very useful research tool for studying bird communications, providing detailed spatial information on territorial birdsong, which may help decipher the song function. This Audio-Zoom can be set up in concert halls, Parliaments where we just want to hear only to specific users. This technique can enhance the speech in such a way that speech recognition will also be easy once speech separation is carried out. This is the long term goal of the project.

This project is to work on the “Cocktail Party Problem” which is a “Blind Speech Separation Problem”. The job is to propose a model of IR including early reflections and a reverberant tail, and test the applicability of it with microphones separated by more than one wavelength, and develop an algorithm to separate the sources (speakers).

Faculty Supervisor:

Dr. Peter Driessen

Student:

M. Wasif Khan

Partner:

Discipline:

Engineering - computer / electrical

Sector:

Information and communications technologies

University:

University of Victoria

Program: