Technion
 
         Computer Science Colloquium
 
    DATE on Thursday 27/01/2011
 
Time+Place : Thursday 27/01/2011 14:30 room 337-8 Taub  Bld.
Speaker    : Tal Sobol Shikler NOTE NEW DATE
Affiliation: Department of Industrial Engineering & Management,
             Ben-Gurion University
Host       : Johann Makowsky
Title      : Analysis of affective behaviour: Non-verbal speech -
             Feature extraction, multi-class & multi-label classification,
             and generalisation of the technology
 
Abstract   :
 
Computation of social and affective behaviour is a field of artificial
intelligence and of human-machine interactions/interfaces which aims to
incorporate human behavioural cues and cues of affect (emotions, mental
states, moods and attitudes) into human-computer and human-robot interfaces,
in order to improve their usability and performance. The talk will describe
the development of a new technology for the recognition of affective states
from their non-verbal expressions in speech in real scenarios. The tone of
voice, the voice quality, and other characteristics of non-verbal speech,
allow us to understand when another person calculates out loud, tries to
choose, is unsure, presents genuine laughter or stressed laughter, even if
we do not understand the spoken words. However, speech contains also the
verbal content, cues that reflect the speaker's identity, background, and
the like. This talk describes how the affective features in speech were
isolated and recognized, i.e. the definition of extraction algorithms of
these features and of their temporal characteristics (signal processing).
The affective states are not mutually-exclusive, several affective states
can occur simultaneously and their levels change dynamically over time. In
addition, different features distinguish different pairs of affective
states. The talk therefore describes a multi-class and semi-blind
multi-label inference algorithm (machine learning) which was designed to
infer the levels of co-occurring affective states. As a basis of a new
technology, the capabilities of the system were further examined, including:
Analysis of affective speech during sustained human-computer interaction, in
another language; comparison of the distinguishing capabilities to human
performance (and outperforming); characterisation of over 500(!) different
affective states and of the relations between them. The design allows this
new technology to be generalised to new affective states, speakers,
languages and scenarios with no additional training.
 
Short Bio:
I have BSc and MSc in EE from Tel-Aviv. PhD in Computer Science and
Technology from the Computer Laboratory, University of Cambridge. The PhD
concerned the analysis of affective speech. It was conducted in the Rainbow
Group with the supervision of Porf Peter Robinson. I am a fellow of
Cambridge Overseas Trust and was awarded the AAUW Educational Foundation
International Award. I am currently a post-doc in the Department of
Industrial Engineering and Management in Ben-Gurion University of the Negev
on intelligent human-robot interaction, and in Deutsche Telekom Laboratories
at BGU in which I took part in various projects related to human-computer
interfaces.
 
----------------------------------------------------------
Visit our home page-   <http://www.cs.technion.ac.il/~colloq>
Mon Jan 17 10:43:12 IST 2011
 
---------------------------------------------------------
Technion Math. Net (TECHMATH)
Editor: Michael Cwikel   <techm@math.technion.ac.il> 
Announcement from: Hadas Heier   <heier@cs.technion.ac.il>