RoboHEARD 2027: Robot Audio-Visual Speech Recognition
An ICASSP 2027 satellite challenge to work out who spoke, when and what they said, using a mobile robot's onboard cameras and microphones.
About RoboHEARD 2027: Robot Audio-Visual Speech Recognition
Track 1 is offline audio-visual speech recognition. Track 2 is streaming and only ranks systems with P95 endpointing delay of 2 seconds or less on one RTX 5090. Registration closes 15 November 2026, the test set and leaderboard open 1 December 2026, and the leaderboard freezes 7 December 2026. In each track, 1st/2nd/3rd place win $1,000/$500/$300.
Who can apply
Not restricted on the page; open datasets and pre-trained models only, declared by 15 Nov 2026; enter one or both tracks.
How to get it
- Register via the Google Form by 15 Nov
- Declare the datasets and pre-trained models you use by 15 Nov
- Submit to the leaderboard after 1 Dec and send a technical report by 15 Dec
RoboHEARD 2027: Robot Audio-Visual Speech Recognition — FAQ
Who can apply for RoboHEARD 2027: Robot Audio-Visual Speech Recognition?+
Not restricted on the page; open datasets and pre-trained models only, declared by 15 Nov 2026; enter one or both tracks.
How do I get RoboHEARD 2027: Robot Audio-Visual Speech Recognition?+
1. Register via the Google Form by 15 Nov 2. Declare the datasets and pre-trained models you use by 15 Nov 3. Submit to the leaderboard after 1 Dec and send a technical report by 15 Dec
How much is RoboHEARD 2027: Robot Audio-Visual Speech Recognition worth?+
$3,600 in prizes. An ICASSP 2027 satellite challenge to work out who spoke, when and what they said, using a mobile robot's onboard cameras and microphones.