gms | German Medical Science

Directional benefit of hearing devices interacts with head motion behavior of the user: for fixed beamformers it can only be achieved if the user is facing the target. Steerable beamformers may alleviate this limitation, but they rely on DOA estimation that can be less accurate with head movements. Furthermore, in multi-talker situations, it remains unclear which of the sources belongs to desired sources. Taken together, current beamformers may not improve speech reception in everyday dynamic listening conditions involving movement of listener as well as in standard lab conditions. To overcome this problem, we proposed an algorithm that estimates auditory attention based on the combination of gaze behavior and DOA estimation [1]. Its aim is to support communication and natural movement behavior in dynamic listening conditions better than fixed beamformers. Attention is modeled in two steps: First, the gaze-to-object probability is estimated by classification of gaze direction based on a acoustic scene model. Next, the temporal statistics of the gaze-to-object probability is analyzed. Finally, non-attended sources are attenuated by a spatial filter. In this study, extensions of the model are introduced and evaluated with normal hearing subjects. Results show that a similar benefit was achieved as with conventional directional microphones. As a tendency, a positive effect towards natural head movement was observed.

Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – Projektnummer 352015383 – SFB 1330 B1