Audio-visual cues based attention switching for machine listening

Designing a machine listening system that mimics the human auditory system is an extremely challenging task. CVSSP, funded by EPSRC and Dstl, has developed the prototype of such a system using the techniques of blind source separation (BSS) and audio-visual (AV) signal processing.

This proposal, by bringing in the expertise from School of Psychology and Department of Computing, attempts to equip the system with certain cognitive capabilities e.g. attention switching, as used naturally by humans in real listening environments and explores its potentials for the detection of abnormal conversations in security applications, such as, safety surveillance and automated crime detection.