2018-01038 - Weakly supervised learning for speech recognition

Renewable contract : Oui

Level of qualifications required : PhD or equivalent

Fonction : Post-Doctoral Research Visit

Context

Inria Nancy - Grand Est is seeking a postdoctoral researcher for a new
European (H2020 ICT) collaborative project called COMPRISE. The
successful candidate will be part of the Multispeech team, which
gathers 40 researchers (faculty, postdocs, PhD students) and software
engineers in the field of speech and audio processing, with focus on
deep learning for speech enhancement and speech/speaker
recognition. The team is very international and English is the working
language.

COMPRISE is a 3-year Research and Innovation Action (RIA) aiming at
new cost-effective, multilingual, privacy-driven voice interaction
technology. This will be achieved through research advances in
privacy-driven machine learning, personalized training, automatic data
labeling, and tighter integration of speech and dialog processing with
machine translation. The technology will be based on existing software
toolkits (Kaldi speech-to-text, Platon dialog processing, Tilde
text-to-speech), as well as new software resulting from these research
efforts.

The consortium includes academic and industrial partners in France
(Inria, Netfective Technology), Germany (Ascora, Saarland University),
Latvia (Tilde), and Spain (Rooter).

Assignment

The postdoctoral researcher will work on the design and the validation
of a weakly supervised learning methodology for automatic speech
recognition. He/she will address the following research questions:

    how to best exploit automatically labeled speech data for
    semi-supervised training of acoustic and language models;

    how to exploit domain-specific context information in order to
    improve the labeling accuracy and the resulting models;

    how to generalize this approach to arbitrary domains.

This research is expected to lead both to publications in top venues
and to a practical software demonstrator.

The research and experimentation will be conducted in the Multispeech
team at Inria Nancy, in tight collaboration with the natural language
processing and dialog team at Saarland University led by
Prof. Dietrich Klakow. The successful candidate will have the
opportunity to visit Prof. Klakow's team regularly, in order to
benefit from the complementary scientific environments offered by the
two teams.

The starting date may be anytime from January 1 to June 1, 2019.

Main activities

    Study existing semi-supervised, active learning methods and their
    applicability in the context of the project's architecture and
    requirements

    Study the properties of label noise incurred by automatic
    labelling and seek to reduce it by exploiting domain-specific
    context information

    Design a weakly supervised learning method that leverages these
    improved labels

    Implement algorithms in a library compliant with the other
    developments of the consortium

    Contribute to demonstrating their benefit for three specific use
    cases (smart consumer apps, e-commerce, and e-health).

    Propose guidelines towards generalizing them to arbitrary use
    cases.

    Coordinate with the two other postdoctoral researchers to be
    recruited by Inria as part of this project, who will be working on
    privacy-friendly machine learning and personalized training

Additional activities

    Publish, report, and disseminate results

    Coordinate with related efforts in the team / community

Skills

    PhD in speech and language processing, applied machine learning,
    or a related field

    Good software engineering and programming skills

    Excellent English writing and speaking skills  

Benefits package

    Subsidised catering service
    Partially-reimbursed public transport
    Social security
    Paid leave
    Flexible working hours

Remuneration

From 2653,00¤  brut per month

General Information

    Town/city : Villers-lès-Nancy
    Inria Center : CRI Nancy - Grand Est
    Starting date : 2019-01-01
    Duration of contract : 2 years
    Deadline to apply :	2018-10-19

Contacts

    Inria Team :	MULTISPEECH
    Recruiter :
    Vincent Emmanuel / emmanuel.vincent@inria.fr

About Inria

Inria, the French National Institute for computer science and applied
mathematics, promotes "scientific excellence for technology transfer
and society". Graduates from the world's top universities, Inria's
2,700 employees rise to the challenges of digital sciences. With its
open, agile model, Inria is able to explore original approaches with
its partners in industry and academia and provide an efficient
response to the multidisciplinary and application challenges of the
digital transformation. Inria is the source of many innovations that
add value and create jobs.

Conditions for application

Defence Security :

This position is likely to be situated in a restricted area (ZRR), as
defined in Decree No. 2011-1425 relating to the protection of national
scientific and technical potential (PPST).Authorisation to enter an
area is granted by the director of the unit, following a favourable
Ministerial decision, as defined in the decree of 3 July 2012 relating
to the PPST. An unfavourable Ministerial decision in respect of a
position situated in a ZRR would result in the cancellation of the
appointment.

Recruitment Policy :

As part of its diversity policy, all Inria positions are accessible to
people with disabilities.

Apply online:
https://jobs.inria.fr/public/classic/en/offres/2018-01038