** The Music and Science list is managed by the Institute of Musical Research (www.music.sas.ac.uk) as a bulletin board and discussion forum for researchers working at the shared boundaries of science and music. **

MESSAGE FOLLOWS:



Dear all,

A reminder that tomorrow, Wednesday, 4th April at 3:00pm, Jon Barker will 
present the seminar 'Coupling speech source separation and recognition 
using a `fragment decoding' approach'.

The talk will take place in room 209 in the Electronic Engineering 
building, Queen Mary University of London, Mile End Road, London E1 4NS. 
Directions on how to access the building can be found at 
http://www.eecs.qmul.ac.uk/about/campus-map.php. If you are coming from 
outside Queen Mary, please let me know, so I can make sure no-one is stuck 
outside the doors. Details of future seminars can be found at 
http://www.eecs.qmul.ac.uk/newsevents/researchgroupevents.php?i=12. All 
are welcome to attend. For those unable to do so, a video recording of the 
seminar will be made available online after a few days.

If you wish to be added to / removed from our mailing list, please send me 
an email and I'll be happy to do so.


Wednesday's seminar (4th April, 3:00pm):

Title:

Coupling speech source separation and recognition using a `fragment 
decoding' approach

Speaker:

Jon Barker

Abstract:

Distant microphone speech recognition presents many challenges. On of the 
chief difficulties is contending with complex multisource noise 
backgrounds. Typical backgrounds are composed of multiple competing sound 
sources, the number of which is generally unknown and whose activity level 
may be changing unpredictably over time. The talk will present approaches 
to tackling this problem that are being developed at Sheffield by the 
EPSRC project CHiME (http://www.dcs.shef.ac.uk/spandh/chime/).

The talk will first present the PASCAL `CHiME' Speech Separation and 
Recognition 
challenge (http://www.dcs.shef.ac.uk/spandh/chime/challenge.html). This 
task employs noise backgrounds from a dataset of over 40 hours of binaural 
audio recorded in various rooms of a busy family home. Real room impulse 
responses have been used to add utterances from a separate speech 
recognition corpus into these multisource environments to provide a 
controlled and yet semi-realistic recognition task. 

Having presented the challenge the talk will motivate our particular 
approach to the solution: speech fragment decoding (SFD). This technique, 
inspired by ideas from Bregman's Auditory Scene Analysis account 
of  perception, exploits monaural and binaural cues for sound source 
separation to locate sound source `fragments' which are then `stitched 
together' using temporal sequence knowledge represented by traditional 
statistical models of speech.  CHiME challenge results will be presented 
along with a discussion of limitations of the technique and directions for 
future research.


Bio:

Jon Barker obtained a degree in Electrical and Information Sciences from 
the University of Cambridge, followed by a Ph.D. in Computer Science from 
the University of Sheffield in 1998. After graduating he spent a year at 
the Institut Communication Parlee, Grenoble, studying audio-visual speech 
perception before returning to the Speech and Hearing Group at Sheffield 
where he is now a Senior Lecturer. Over the years he has spent time as a 
visiting researcher at ICSI, IDIAP and Columbia University. His research 
interests include modelling speech perception in real environments, 
noise-robust speech recognition and computational hearing for robotics. 
Much of his recent research has concerned the attempt to design robust 
speech technology using ideas inspired by our limited understanding of how 
humans process speech in noisy acoustic environments.



Future C4DM seminars (Seminar details tbc):

Laurent Simon - INRIA Rennes Bretagne Atlantique, France
Thu 19th April 2012

Wenwu Wang - University of Surrey
Wed 25th April 2012

Martyn Davies - Six Two Productions
Wed 9th May 2012





--
Peter Foster
Postgraduate Research Student
Room 104, Electronic Engineering Bldg
Centre for Digital Music
Queen Mary, University of London
Mile End Road, London E1 4NS, UK
email: [log in to unmask]