Blind Estimation of Head Rotations From Binaural Recordings

Abstract:
Binaural recordings measured at the human head inherently include head rotations. Knowledge of the current head rotation can be helpful for applications like speech enhancement in hearing aids, as it provides information about changes in the spatial cues. However, the head rotations may be unknown in advance. In this contribution, we propose two algorithms that utilize binaural directionof- arrival (DoA) estimation techniques to blindly estimate head rotation from binaural recordings. We evaluate both algorithms using real binaural recordings and analyze how the cone-of-confusion (CoC) problem affects the estimation performance.