The Effect of Deep Neural Network Implementation on Speech Recognition, Listening Effort, and Sound Quality in Older Adults With Mild to Moderately Severe Hearing Loss.
Paula Folkeard, Vahid Ashkanichenarlogh, Mohamed Rahme and 6 others
DNN noise reduction combined with beamforming in a Phonak hearing aid outperformed DNN alone, traditional noise reduction with beamforming, and no processing across speech recognition, clarity, total impression, listening effort, and background noise awareness in 20 older adults with mild to moderately severe hearing loss.
Key findings
1Beam + DNN outperformed all three other hearing aid programs across all five outcome measures: speech recognition, clarity, total impression, listening effort, and background noise awareness.
2The advantage of Beam + DNN over the traditional noise-reduction-plus-beamforming program was most apparent in speech-shaped noise and when speech came from the side. In multitalker babble from the front, directionality was the main driver and the DNN added less.
3In speech-shaped noise from the side, speech recognition scores were near ceiling (89.4% to 95.1% correct across programs), so the subjective ratings captured differences that the recognition scores did not.
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
Speech recognition in speech-shaped noise was near ceiling (89.4% to 95.1% correct), so the recognition scores may not capture the full range of differences between programs. The subjective ratings did not show this ceiling. The two beamforming conditions used different types of beamforming (monaural adaptive in the DNN condition, binaural in the traditional noise-reduction condition), so the comparison is not purely about the noise-reduction algorithm. The background noise awareness interaction analysis was underpowered for the program-by-azimuth effect. All testing was in a sound-treated booth with controlled noise sources; the authors note that whether these results hold in real-world environments would require further study. Twenty participants at a single site, all using the same Phonak model, limits generalisability to other hearing aid brands, DNN implementations, and hearing loss configurations.
Declared interests
Sonova (parent company of Phonak) provided research funding and the hearing devices used in the study. A Mitacs Accelerate grant (IT29442) also contributed. Three of the authors (V.K., J.Q., B.S.) are employees of Sonova AG.
The easy way to misread this
Do not read this as evidence that DNN hearing aids are broadly superior to traditional noise reduction. This is one DNN implementation on one Phonak model, funded by Sonova (Phonak's parent company), with three authors employed by Sonova. In speech-shaped noise from the side, recognition scores were near ceiling (89.4% to 95.1% correct), so the recognition differences in that condition are small even though the subjective ratings showed larger gaps.
Summarised by AI from the full paper, without a clinician reviewing it. Check it against the source before it changes what you do. Read it on PubMed →