[Comp-neuro] Fully funded PhD position at the Laboratoire des Neuroscience Cognitive and Group for Neural Theory, ENS, Paris, France

Boris boris.gutkin at gmail.com
Fri Oct 1 15:01:09 CEST 2010

A fully funded PhD is available in the Laboratoire de Neuroscience  
Cognitives (Departement des Etudes Cognitives, ENS, Paris) in  
computational models of speech perception.

The interdisciplinary PhD will be co-supervised by Anne-Lise Giraud  
(neuroscience, Auditory Language Group http://www.giraud.ens.fr/ ) and  
Boris Gutkin (modelling, Group for Neural Theory http:// 
www.gnt.ens.fr/ ).  A more detailed description is given below.

We encourage applications from students with quantitive backgrounds  
(math, physics, engineering) interested in neuroscience training and/ 
or students with a strong neuroscience background and quantitive  
training interested in computational modeling.

Interested individuals should contact: Alexandre Hayfil (alexandre.hyafil at gmail.com 
); Anne-Lise Giraud (Anne-Lise.Giraud at ens.fr) or Boris Gutkin (boris.gutkin at ens.fr 

Project description:

Laboratoire de Neurosciences Cognitives, Département d’Etudes Cognitives
Auditory Langauge Group/ Group for Neural Theory

Ecole Normale Supérieure, Paris

The neural mechanisms underpinning such a complex function as speech  
comprehension remain to a large extent unknown. Speech neural networks  
have been identified within the brain but it is not clear how the  
continous acoustic input is segmented and subsequently categorized  
into units of different time scales: phonemes, syllables, words, and  
ultimately sentences to which semantic significance can be attached.  
Actually, it is not even clear how relevant these different categories  
are to brain processing.

A wealth of data suggests that speech processing at each level (from  
phonemes to words) is paralelled by synchronisation of neural  
responses at a corresponding frequency. Moreover, these oscillations  
seem to be nested into each other, so that the lower frequency  
oscillator controls the amplitude of higher frequency. A recent  
phenomenological model ties up all these elements by proposing that  
speech processing is controlled by a cascade of neural oscillators  
(from 5-10 Hz to 60-120 Hz) that chunk the acoustic signal into bits  
of the relevant size. The oscillators would be phase-locked to the  
signal and thus allow appropriate discretization and categorization,  
thanks to nesting mechanisms.

The properties of this model need to be clarified and put together  
into a computationnal model in order to prove its operability and  
provide further tests for neural investigation. At first, though  
single-frequency neural oscillators are very well documented, a  
plausible model is yet to be proposed showing how low frequency  
oscillations can control high frequency oscillations. Second, the  
model has so far been only tested on rudimentary signal on a single  
time scale. Building a full model with interacting oscillators being  
able to deal with natural speech and setting the pathway towards  
access to the lexicon is thus the final objective of this work.

This project will be undertaken by a joint team involving the hired  
PhD student and a postdoctorate of the lab (Alexandre Hyafil). It will  
be supervised both by Anne-Lise Giraud for the functional components  
and Boris Gutkin for the modelling aspects. This work also involved a  
collaboration with Oded Ghitza (Center for Biodynamics, Boston  
University). Computational modelling will be the primary tool. It is  
meant to bring together elements taken from a number of scientific  
domains, including neurophysiology, neuroimaging, psycholinguistics,  
phonetics and automatic speech recognition.


Shamir M, Ghitza O, Epstein S, Kopell N. Representation of time- 
varying stimuli by a network exhibiting oscillations on a faster time  
scale. PLoS computational biology. PLoS Comput Biol. 2009 May;5(5).

Ghitza O. Speech decoding guided by cascading oscillators locked to  
the sensory-input rhythm: towards predicting intelligibility of time- 
compressed speech with insertions of silence. Submitted


More information about the Comp-neuro mailing list