Self-organized adaptive processor for solving
the cocktail party problem
14:00-15:00 Monday Mar 27, 2006
IMM BUILDING 321 ROOM 053
Prof. Simon Haykin
Director, Adaptive Systems Laboratory
Hamilton, ON Canada
In this lecture, I first highlight the important aspects of the
cocktail party problem (CCP). These highlights set the stage of the
description of a self-organized adaptive processor for solving the CPP.
Design of the processor is motivated by neurobiological considerations
of the human auditory system. Specifically, the processor is binaural,
consisting of three functional blocks:
1. Pre-processor for extracting four acoustic cues: Interaural time
difference (ITD), Interaural intensity difference (IID), Onset and Pitch.
The cues provide the basis for separating the target signal from
background noise. Moreover, the two spatial cues provide the means
for focusing attention on the target signal.
2. Fusion scheme for combining the cues considered to belong to the
3. Adaptive smoother for suppressing musical noise, a function that is
achieved by feeding back the extracted
spatial cues to the input of the smoother.
The lecture will finish by presenting audio recordings under different
signal-to-noise ratios and varying reverberant conditions.
The recordings will demonstrate the impressive
performance of the new processor.
Simon Haykin is a pioneer in signal processing and machine learning,
author of several widely used textbooks, for more information:
Prof. Lars Kai Hansen
Informatics and Mathematical Modelling
Technical University of Denmark B321
DK-2800 Kgs. Lyngby - DENMARK
Tel: (+45) 45253889 Fax: (+45) 45872599