13th Speech in Noise Workshop, 20-21 January 2022, Virtual Conference 13th Speech in Noise Workshop, 20-21 January 2022, Virtual Conference

P29 Mapping of adaptive SNR enhancement in modern hearing aids: Effects of compression and choice of noise reduction strategy

Marie Frederikke Garnæs
Centre for Applied Audiology Research, Oticon A/S, Smørum, Denmark

Sébastien Santurette
Centre for Applied Audiology Research, Oticon A/S, Smørum, Denmark | Department of Health Technology, Technical University of Denmark, Kgs. Lyngby, Denmark

(a) Presenting

For people with sensorineural hearing loss, the ability to understand speech in the presence of noise can be significantly degraded. In noisy sound environments, the low signal-to-noise ratio (SNR) causes reduced speech intelligibility and increased listening effort. While modern hearing aid (HA) technology accommodates this by enhancing SNR in aided listening, individual listeners may have different needs for SNR improvement in a given listening situation. Modern premium HAs can be programmed to obtain a wide range of SNR improvement patterns to accommodate a multitude of individual hearing abilities. This study investigated the amount of SNR enhancement provided by noise reduction strategies consisting of a combination of beamforming (BF) and postfiltering (PF) algorithms. For this purpose, the SNR at the output of recent premium HAs was measured across a wide range of input SNRs and BF+PF activation settings, using the Hagerman & Olofsson phase inversion technique in an ecologically valid speech-test setup. The aims were to document the achievable SNR enhancement of current algorithms, to compare the SNR benefit of traditional vs deep-neural-network (DNN) based PF algorithms, and to estimate the effects of wide-dynamic range compression on output SNR. The results confirmed that a wide range of SNR enhancement patterns could be obtained depending on the desired level of help in noise provided to the user. SNR enhancements were more prominent with decreasing input SNRs and the tested BF+PF algorithms could provide up to 10 dB SNR enhancement at low input SNRs. The DNN-based PF was shown to provide a relatively stable SNR enhancement of up to 2 dB across all input SNRs on top of the enhancement provided by the BF, largely improving the enhancement provided by the PF relative to the traditional PF approach. In accordance with previous literature, applying compression decreased output SNR at positive input SNRs, which is an unintended consequence of the feature. These SNR-reducing effects of compression typically increased with higher BF+PF activation. Overall, these technical measurements demonstrate the wide range of possibilities for individualization of advanced noise reduction settings available in the latest HA technology.

Last modified 2022-01-24 16:11:02