Upcoming SPEAR Challenge: SPeech Enhancement for Augmented Reality (SPEAR) challenge

Augmented reality devices of the future will likely fuse sensor data from several modalities, allowing multichannel speech enhancement algorithms to exploit, for example, head orientation and accurately estimated source directions. Augmented hearing therefore offers the potential for easier communication in noisy environments, such as restaurants and bars.

The SPEAR challenge offers the opportunity for researchers to benchmark existing speech enhancement algorithms and try out new ideas in the context of head-worn microphone arrays where positional information is available to the algorithm. Using recorded and simulated data, participants will obtain their best estimate of a binaural target signal from a given microphone array where real time information of the wearer head rotation and of the target direction of arrival is available. Enhanced signals will be evaluated through metric calculation and crowdsourced listening tests. To help get started a baseline algorithm is provided.

The challenge is being run by the Speech and Audio Processing lab at Imperial College London in partnership with Reality Labs Research (Meta). More information can be found on the website along with all the freely available data and tools. We encourage people to register for the latest information and to follow the SPEAR challenge on Twitter.

Important Dates

• September 2022:
  
  • Start of the challenge.
  • Release of Train and Dev datasets.
  • Release of baseline and tools.
• 4^th of January 2023:
  
  • Evaluation data released.
• 25^th of January 2023:
  
  • Submission of Evaluation data.
• February 2023:
  
  •  Analysis of submitted enhanced audio using metrics calculation and crowdsourced listening tests.
• March 2023:
  
  • Workshop presenting results and findings.

For more information, email, spear.challenge@gmail.com.