SLPOtherTrends in hearing2018

Use of a Deep Recurrent Neural Network to Reduce Wind Noise: Effects on Judged Speech Intelligibility and Sound Quality.

Mahmoud Keshavarzi, Tobias Goehring, Justin Zakis and 2 others

PMID 29708061

WHAT IT FOUND

Listeners preferred hearing aid recordings with wind noise reduced by a computer algorithm over unprocessed wind-noise recordings, but not consistently over simple high-pass filtering.

Preferences were small and lab-based, not tested in real hearing aids.

Key findings

01The recurrent neural network was significantly preferred over unprocessed noisy speech for intelligibility and sound quality.

02High-pass filtering was not significantly preferred over unprocessed noisy speech.

03The recurrent neural network and high-pass filtering were not significantly different except for sound quality in hearing-impaired participants.

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.

Already have one?

What it does not show

Only 18 participants were tested, and they listened to recordings rather than wearing the hearing aids. The wind speed was 3 m/s; higher wind speeds were not used because speech-to-wind-noise ratio was very low. Only three wind and speech directions were tested. The high-pass filter was steeper than typical hearing aid filters, so it may not match everyday device processing. Participants reported audible artifacts with the recurrent neural network, such as gurgling and fluctuating noise. The study did not test the algorithm inside a hearing aid or everyday communication outcomes.

Declared interests

The work was funded by the Allen Trust, Action on Hearing Loss, and the Engineering and Physical Sciences Research Council.

The easy way to misread this

Do not conclude that the recurrent neural network is a clinically useful hearing aid wind-noise reduction method. The study used lab recordings, small preferences, and no real hearing aid use. High-pass filtering was not clearly worse.

Read it on PubMed →