Evaluation of deep marginal feedback cancellation for hearing aids using speech and music.
Chengshi Zheng, Chenyang Xu, Meihuang Wang and 2 others
PMID 37551089WHAT IT FOUND
Deep learning feedback cancellation worked well for speech but degraded music quality, especially when trained only on speech.
Combining it with traditional adaptive feedback cancellation improved stability and preference for music, but the deep model alone was not sufficient.
Key findings
01DeepMFC(1), trained on speech, provided the highest additional stable gain among individual approaches for speech, up to 14 dB.
02For music, DeepMFC(1) failed to increase maximum stable gain and degraded sound quality even in stable states.
03Listeners preferred the combination of PEM-AFC and DeepMFC(3) over either method alone for music at negative gain margins.
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 simulated hearing loss profiles (mild and moderate) and normal-hearing listeners for the subjective test, not actual hearing-aid users. The feedback paths were measured but the system was simulated; real-world acoustic variability and user movement were not fully captured. The deep learning models had limited parameter sizes, which may have restricted their ability to handle the complex spectro-temporal structure of music. Only speech and music were evaluated; environmental sounds were not tested.
Declared interests
The authors declared no potential conflicts of interest. The work was supported by non-U.S. government funding.
The easy way to misread this
Do not assume that deep learning feedback cancellation will improve music quality for your patients. The study found that models trained primarily on speech degraded music significantly, and even models trained on music performed worse than traditional adaptive cancellation at stable gains.