[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Announcing the 2nd CHiME Speech Separation and Recognition Challenge



[with apologies for cross-postings]

Dear Auditory list,

The CHiME challenge has been designed for the speech processing, source
separation and machine learning communities in particular, as an
opportunity to evaluate signal enhancement and recognition algorithms
in the hot field of distant-microphone speech recognition. Many teams
from different communities participated in the first edition and we hope
even more of you will participate this year!

In case you are not an expert in speech processing, a complete set of
speech recognition tools based on HTK are provided so you may focus on
signal enhancement and simply run the baseline recogniser to derive speech
transcripts in the end.

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

  2nd CHiME Speech Separation and Recognition Challenge
          Supported by IEEE Technical Committees

                Deadline: January 15, 2013
        Workshop: June 1, 2013, Vancouver, Canada


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


Following the success of the 1st PASCAL CHiME Speech Separation and
Recognition Challenge, we are happy to announce a new challenge
dedicated to speech recognition in real-world reverberant, noisy conditions,
that will culminate in a dedicated satellite workshop of ICASSP 2013.

The challenge is supported by several IEEE Technical Committees and by
an Industrial Board.


FEATURED TASKS

The challenge consists of recognising distant-microphone speech mixed in
two-channel nonstationary noise recorded over a period of several weeks
in a real family house. Entrants may address either one or both of the
following tracks:

Medium vocabulary track: WSJ 5k sentences uttered by a static speaker

Small vocababulary track: simpler commands but small head movements


TO ENTER

You will find everything you need to get started (and even more) on the
challenge website:
- a full description of the challenge,
- clean, reverberated and multi-condition training and development data,
- baseline training, decoding and scoring software tools based on HTK.

Submission consists of a 2- to 8-page paper describing your system and
reporting its performance on the development and the test set. In
addition, you are welcome to submit an earlier paper to ICASSP 2013,
which will tentatively be grouped with other papers into a dedicated
session.

Any approach is welcome, whether emerging or established.

If you are interested in participating, please email us so we can
monitor interest and send you further updates about the challenge.


BEST CHALLENGE PAPER AWARD

The best challenge paper will distinguished by an award from the
Industrial Board.


IMPORTANT DATES

July 2012          Launch
October 2012       Test set release
January 15, 2013   Challenge & workshop submission deadline
February 18, 2013  Paper notification & release of the challenge results
June 1, 2013       ICASSP satellite workshop


INDUSTRIAL BOARD

Masami Akamine, Toshiba
Carlos Avendano, Audience
Li Deng, Microsoft
Erik McDermott, Google
Gautham Mysore, Adobe
Atsushi Nakamura, NTT
Peder A. Olsen, IBM
Trausti Thormundsson, Conexant
Daniel Willett, Nuance


WORKSHOP SPONSORS

Conexant Systems Inc.
Audience Inc.
Mitsubishi Electric Corp.


ORGANISERS

Emmanuel Vincent, INRIA
Jon Barker, University of Sheffield
Shinji Watanabe & Jonathan Le Roux, MERL
Francesco Nesta & Marco Matassoni, FBK-IRST

-- 
Dr. Jon Barker, Department of Computer Science,
University of Sheffield, Sheffield,  S1 4DP, UK
Phone: +44-(0)114-22 21824 FAX: +44-(0)114-222 1810
Email: j.barker@xxxxxxxxxxxxxx  http://www.dcs.shef.ac.uk/~jon

-- 
Dr. Jon Barker, Department of Computer Science,
University of Sheffield, Sheffield,  S1 4DP, UK
Phone: +44-(0)114-22 21824 FAX: +44-(0)114-222 1810
Email: j.barker@xxxxxxxxxxxxxx  http://www.dcs.shef.ac.uk/~jon