ResearchBib Share Your Research, Maximize Your Social Impacts
Sign for Notice Everyday Sign up >> Login

CHiME 2015 - 3rd CHiME Speech Separation and Recognition Challenge

Date2015-12-13 - 2015-12-17

Deadline2015-02-20

VenueScottsdale, Arizona, USA - United States USA - United States

Keywords

Websitehttps://spandh.dcs.shef.ac.uk/chime_challenge

Topics/Call fo Papers

Following the success of the 2011 and 2013 CHiME challenges it gives us great pleasure to pre-announce the 3rd CHiME Speech Separation and Recognition Challenge (CHiME-3)
CHiME-3 will be an official IEEE ASRU 2015 Challenge Task. Participants will be invited to submit CHiME-3 papers to the ASRU workshop to be held in Scottsdale, Arizona 13-17 December. Papers will be presented at a Special Session.
THE TASK
The CHiME-3 scenario will be ASR for a multi-microphone tablet device in everyday, noisy environments. It will represent a significant step forward in terms of both realism and difficulty with respect to the previous CHiME challenges.
The challenge will feature:
- 6-channel microphone array data,
- real acoustic mixing, i.e. talkers speaking in challenging noisy environments,
- varied noise settings including cafe, street junction, public transport.
To maintain compatibility with the 2nd CHiME challenge, the new challenge will re-use the WSJ evaluation framework. Utterances will be provided embedded in continuous audio with ground truth VAD annotations.
MATERIALS
At time of launch in February we will provide:
- a development test set, recorded by 4 US talkers across 4 noise environments,
- a real training set, comprised of 2000 utterances spoken by 4 US talkers in noisy environments plus several hours of noise background per environment,
- tools for generating a simulated training set by remixing WSJ and background audio with impulse responses estimated from the real data,
- a reference speech enhancement system and a state-of-the-art DNN-based Kaldi ASR system.
As with previous CHiME challenges we invite participation from both the signal processing and the speech recognition communities. To support teams who lack access to the necessary GPU infrastructure required to run the evaluation system, we will offer 'remote evaluation' as a service.
If you are considering participating please email chimechallenge-AT-gmail.com and you will be added to the email list for receiving further updates.
IMPORTANT DATES
Feb 20, 2015 -- Launch - Training data + dev data release
May 15, 2015 -- Test set released
July 15, 2015 -- Challenge paper submission deadline
September 11, 2015 -- Paper notification & release of CHiME-3 results
December 13-17, 2015 -- ASRU Workshop
ORGANISERS
Jon Barker, University of Sheffield, j.p.barker-AT-sheffield.ac.uk
Ricard Marxer, University of Sheffield, r.marxer-AT-sheffield.ac.uk
Emmanuel Vincent, Inria, emmanuel.vincent-AT-inria.fr
Shinji Watanabe, MERL, watanabe-AT-merl.com

Last modified: 2014-12-17 22:22:55