THE SPEAR CHALLENGE - REVIEW OF RESULTS

Verbal communication can be challenging in the presence of acoustic noise. To tackle this problem, microphone arrays coupled with numerous processing methods have been studied in the past few decades. Recent interest in Augmented Reality (AR) applications gives rise to head-worn microphone arrays. This highlights additional important aspects such as motion of the capture device with respect to the scene, the need to preserve spatial characteristics in the processed sound, and the tightened constraints on latency and computational budgets. The SPeech Enhancement for Augmented Reality (SPEAR) Challenge, endorsed by the IEEE Challenges and Data Collection initiative, was organized in order to further ignite interest in this important problem and to obtain a better sense of the remaining technological gaps. The challenge is based on an adaptation of the recently published EasyCom dataset that contains noisy conversation recordings from a glasses form-factor AR device with 6 microphones along with the positional information and additional labeled modalities. A competitive evaluation of the challenge entrant algorithms was carried out by using both objective metrics and subjective listening tests. The current contribution is focused on providing an overview of the SPEAR challenge and highlighting some of the most important findings and outcomes.