A postdoc position in unsupervised speech recognition is available.
It is part of a two-year experimental/computational phonology research
project on phonological inventories (GEOMPHON : PI: Ewan Dunbar). The
project aims to explain why the sound inventories of human languages show
typological tendencies toward being phonetically coherent in various
geometrically definable ways; a classical example of this is “economy,”
whereby the contrasts found in inventories appear to exploit surprisingly
small-dimensional feature sub-spaces; others have been discovered more
recently (Dunbar and Dupoux 2016, Frontiers in Psychology).
Perceptual and artificial language experiments will test explanations of
these tendencies. The job of the postdoc will be to explore unsupervised
speech representation learning, constructing models inspired by these
typological tendencies which will be evaluated by matching their behaviour
against the results of new data from the perceptual experiments.
The main host laboratory, the Laboratoire de Linguistique Formelle (LLF),
is an interdisciplinary research laboratory linked to the Department of
Linguistics (UFR Linguistique) at the Université Paris Diderot. The LLF
brings together researchers from computational, experimental, theoretical,
and descriptive linguistics. The project is a collaboration with the
Laboratoire des Sciences Cognitives et Psycholinguistique (LSCP) at the
École Normale Supérieure (affiliated project members : Emmanuel Dupoux,
LLF and LSCP are high quality, complementary, environments for the
development of researchers specializing in computational or experimental
approaches to speech, between them having top-tier expertise in
psycholinguistics (developmental and adult), phonetics and phonology,
natural language processing, unsupervised speech recognition, language
typology, and corpora, both textual and speech. Adjoining labs have
expertise and empirical resources in the neural basis of audition. The
CoML team (INRIA Cognitive Machine Learning group:
http://www.syntheticlearner.net/), of which the postdoc would be a member,
organizes the biannual ZeroSpeech Unsupervised Speech Challenge.
Université Paris Diderot
The ideal candidate for the unsupervised speech recognition will have a
large amount of experience in machine learning applied to speech, and in
constructing cognitively meaningful model evaluations.
The postdoc will have defended their dissertation by the start date. The
start date is somewhat flexible. The submission deadline is strict and
urgent, since visa processing can take some time for non-EU applicants.
Duration: 12 months, with possible extension for an additional 3 months.
About Université Denis Diderot (Paris)
Universit Paris Diderot offers a wide range of degrees in the Humanities, Medicine and the Sciences.