[LUG.ro] Fwd: Connectionists: Machine learning and speech recogniton, PhD and postdoc at LIMSI, Paris

Horacio Castellini lugro@lugro.org.ar
Mon, 4 Jul 2005 16:03:05 -0300


Reenvío este mensaje, por si alguien de la lista le interese hacer un 
doctorado en el tema sobre la conferencia que se dió en el CEC, por supuesto 
es para LCC ó LCC del último año...

Subject: Connectionists: Machine learning and speech recogniton, PhD and
postdoc at LIMSI, Paris
Date: Thu, 30 Jun 2005 14:03:40 +0200
From: Holger Schwenk <schwenk@limsi.fr>
To: connectionists@cs.cmu.edu
CC: Holger Schwenk <Holger.Schwenk@limsi.fr>


New learning algorithms for large vocabulary Speech Recognition
PhD and postdoc positions at LIMSI-CNRS, Orsay, France

The speech processing group at LIMSI-CNRS in Orsay (near Paris) has a long
experience in conducting research in all aspects of speech processing. We
have developed large vocabulary speech recognizers for broadcast news and
conversation speech in several languages (English, French, German, Spanish,
Chinese, Arabic, ...).  We are currently involved in several national and
international projects, in particular the integrated European projects
TC-STAR and CHIL.

Funding for a 3 year PhD and a 1 year position (renewable) is available.
Support for conference travel is provided.  We are in particular interested
in candidates working on the application of new promising learning algorithms
from the general machine learning community to large vocabulary speech
recognition.

When large amounts of acoustic training data are available (>500h), it seems
suboptimal to train the acoustic models directly on all the data.  We want
to
explore alternative ways to take better advantage of the available
resources,
e.g. adaptive data selection, resampling techniques or mixture models. It is
also common to combine several speech recognizers using system combination
(rover and consensus network combination). These multiple systems are
usually
build in an ad-hoc way and it would be better to train explicitly systems
that
combine well.  This could be done by boosting-like methods that construct
sequentially classifiers in function of the errors of the preceding ones.
Another topic of interest are continuous space language models.  We want to
investigate different alternative probability estimators and techniques for
unsupervised language model adaptation.  A large Linux cluster is available
to
support compute extensive research.

The candidate for the PhD position should hold a master in Computer Science,
Electric Engineering or equivalent with experience in the following areas:
large vocabulary continuous speech recognition, machine learning, neural
networks and statistics.  Good programming skills in C and working
experience
on Linux machines is a necessary condition.  The candidate for the postdoc
position is expected to have an established research record in the same
areas.
The positions are available immediately.

Application should be sent to Holger Schwenk (schwenk@limsi.fr) or Jean-Luc
Gauvain (gauvain@limsi.fr) with a detailed CV, list of followed classes (for
the PhD position), list of publications (for the postdoc position), and
letters
of recommendation or name of references.

-------------------------------------------------------