Cosima A. Ermert1, Lu Xia2, Brian Man Kai Loong2, Janina Fels1, Sébastien Santurette2,3
1Institute for Hearing Technology and Acoustics, RWTH Aachen University, Aachen, Germany
2Centre for Applied Audiology Research, Oticon A/S, Smørum, Denmark
3Department of Health Technology, Technical University of Denmark, Kgs. Lyngby, Denmark
Focusing on target signals in noisy environments is a well-known challenge for hearing-impaired listeners. Therefore, a main goal of hearing aid (HA) technology is to enhance target signals and attenuate background noise. An established objective measure to quantify this contrast achieved by the HA is the output signal-to-noise ratio (SNR), commonly estimated using the phase-inversion method (Hagerman and Olofsson, 2004). Output SNR measurements are typically performed in artificial lab setups where target and noise signals are played from discrete directions. However, while such setups can be controlled and modified precisely, they do not adequately represent everyday listening situations. This study investigated the potential and limitations of bringing ecological validity to SNR measurements by using ambisonic reproduction of real sound scenarios as stimuli. Measurements with HAs using different directionality and noise-reduction strategies were performed in multiple pre-recorded 3D scenes. While this allowed a comparison of how the different strategies handled real-life background sounds against specific target sounds, it also introduced challenges to the phase-inversion method. Modern HAs rely increasingly on non-linear processing to adapt to different scenarios, while the phase-inversion method is most accurate under linear processing conditions. Thus, some HA features used for, e.g., feedback management, may induce errors to SNR measurements. The present results underlined the importance of defining an acceptance criterion for the error signal in output SNR measurements. They provided insights on how HA settings must be chosen to eliminate distortions while still measuring the devices at their full potential. When this is considered, measuring HA output signals under realistic conditions can have further useful applications for objective or subjective evaluation. As the increasing complexity of signal processing algorithms is challenging established methods for assessing HAs, it is necessary to discuss which considerations must be made to ensure ecologically valid and future-proof evaluation techniques.
Acknowledgements: The authors would like to thank Jens-Christian Britze Kijne and Boris Søndersted for lab setup, Alexandros Rompelakis for ambisonic sound environment rendering, Asger Heidemann Andersen and Jacob Aderhold for guidance on SNR measurements and post analysis.