ISCApad Archive » 2018 » ISCApad #238 » Academic and Industry Notes » Air Traffic Control Speech Recognition Challenge |
ISCApad #238 |
Friday, April 13, 2018 by Chris Wellekens |
Air Traffic Control Speech Recognition Challenge
AIRBUS, in collaboration with IRIT and SAFETY DATA-CFH, is launching a challenge on automatic speech recognition of English Air Traffic Control (ATC) communications to:
Why is it challenging? ATC audio is very noisy, made of non-native speech, with code switching, high speech rate and lots of domain-specific vocabulary -- see https://www.youtube.com/watch?v=sxEfwSNBgNU&t=80s for example.
What are the tasks?
What is the dataset? 40h of transcribed real-life non-native English ATC data + 5h for validation/leader board + 5h for final evaluation (you can use your own additional data)
What is to win? #1 will be invited to speak at the challenge workshop and to visit Airbus facility and simulators. 5 of the best challengers will be retained for a second internal selection aiming to choose Airbus R&T partner for developing future on-board speech-to-text solutions.
Who can participate? Anyone: start-ups, medium & big companies, research labs. You can participate in the challenge even if you are not considering a partnership with AIRBUS or if you are planning to do only one of the tasks.
Important dates Declaration of intent: Now, write to atc-challenge@airbus.com Registration open: Mon 5th February, via AIRBUS challenge platform https://airbusaigym.nova.airbusdefenceandspace.com/ Training data available + leader board: Mon 5th March - Sun 29th April (8 weeks) Release of evaluation data: Wed 2ndMay Results submission: Wed 9thMay Final workshop: between Mon 18th and Fri 22nd June (date to be confirmed) Special session at international conference for publication: pending approval (submitted)
More questions? Post to the challenge forum or write to atc-challenge@airbus.com.
|
Back | Top |