Objective Evaluation of a Deep Learning-Based Noise Reduction Algorithm for Hearing Aids Under Diverse Fitting and Listening Conditions.
Vahid Ashkanichenarlogh, Paula Folkeard, Susan Scollie and 2 others
PMID 41289071WHAT IT FOUND
Deep learning noise reduction in hearing aids significantly improved predicted speech intelligibility compared to traditional methods, especially when combined with beamforming.
These benefits were strongest in steady noise and for side-coming speech, but the study used computer models rather than human listeners.
Key findings
01The deep learning noise reduction algorithm, particularly when combined with beamforming, consistently outperformed traditional adaptive filtering and beamforming methods in predicted speech intelligibility scores.
02The deep learning algorithm provided significant benefits for speech coming from the side, whereas traditional beamforming strategies actually performed worse than no processing in this condition.
03The study relied entirely on objective computer metrics to estimate intelligibility and quality, with no data from actual human listeners.
STILL TO COME
How it was doneWhat they foundWhat it means for SLPs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The study used a mannequin and computer metrics to predict intelligibility and quality, not actual human listeners. Objective metrics may not reliably predict how humans perceive speech processed by deep learning algorithms. The study did not have an adequate sample of simulated listeners to statistically analyze the effects of different audiogram severities. The non-intrusive pMOS metric does not account for hearing loss effects. Funding and authorship conflicts of interest exist, with one author employed by the hearing aid manufacturer and research funded by the manufacturer.
Declared interests
The research was funded by Sonova (a hearing aid manufacturer) and Mitacs Accelerate. One author is an employee of Sonova AG.
The easy way to misread this
Do not interpret the improved computer-predicted intelligibility scores as proof that patients will understand speech better. The study used a mannequin and objective metrics, not human listeners, and the authors explicitly warn that these metrics may not reliably predict actual human perception for deep learning processing.