Authors: Melika Kianian¹, Hendrik Kayser¹, Volker Hohmann¹
¹Auditory Signal Processing & Hearing Devices and Cluster of Excellence “Hearing4all.connects”, Department of Medical Physics and Acoustics, University of Oldenburg, Oldenburg, Germany
Background: A critical aspect of hearing aids is speech enhancement in challenging multi-talker environments with background noise. Previous studies using oracle knowledge with imposed estimation errors supported the use of multi-beam minimum variance distortionless response (MVDR) beamforming. This work advances that approach by implementing a participant-in-the-loop operation using real-time direction-of-arrival (DOA) estimation.
Method: Real-time DOA estimation was integrated into openMHA to steer three processing conditions: no beamformer, single-beam MVDR, and multi-beam MVDR. TASCAR was used to render an acoustic scene with two target talkers at ±30° in diffuse cafeteria noise. Pilot measurements were conducted with three participants using the German Matrix Sentence Test, implemented as two interleaved adaptive tracks. Performance was evaluated using Speech Reception Threshold (SRT80) and offline SNR improvement analyses.
Results: The real-time DOA-guided MVDR framework was successfully implemented. Pilot results showed the best behavioral outcome for multi-beam MVDR compared with both single-beam MVDR and no beamforming. Logged data facilitated post-analysis of DOA steering behavior and subject performance.
Conclusion: This work demonstrates the suitability of the proposed framework for participant-in-the-loop evaluation of hearing aid signal enhancement. The results support multi-beam MVDR as a promising approach for robust speech enhancement in realistic multi-talker listening conditions.
