Transcript Multichannel processing

ACG Progress Meeting
31 January 2008
Multichannel auditory processing in
determining gaze direction
Problem statement
› How to make a robot (iCat or observation camera)
gaze at an attention requiring event detected through
an auditory event.
Solution ingredients
› Robot control
› Sound recognition
› Sound localization
Approach
› Mixture of two approaches:
• Heuristical
• builds on work in the group
• derived from cognitive science
• Neuromimetic
• builds on my own computational neuroscience
background
Necessary first step:
› Making multichannel auditory processing possible
within cpsp-framework.
Multichannel processing
› Make like mechanism for treating feedforward
dependencies:
• Representations (D.audio, D.E etc.) in role of build
targets
• processSoundFunctions (read(m), calcEdB) in role
of build rules
› Preserve existing signal processing functionality
› First iteration works offline
› Second iteration should work with live data
Demo
Outlook:
› Working with life data
› Heuristical approach
• Apply simple correlation methods on onsets
categorized sound events to detect direction.
› Neuromimetic approach
• Create field of neurons in which different
directions separate into different streams,
mimicking function of the LSO and MSO areas.
Other continuing work
›
›
›
›
›
Inhibitory molding during attention
Carbachol induced oscillations
Topology and firing patterns. (Write)
Improved Carnevale-Hines algorithm. (Rewrite)
Analytic solution of the BESTL model for dendritic
branching. (Write)
› Influence of GABA-ergic development on visual
cortical processing (Write, Remodel?)