Development of a Deep Neural Network for Speeding Up a Model of Loudness for Time-Varying Sounds.
Josef Schlittenlacher, Richard E Turner, Brian C J Moore
PMID 32853098WHAT IT FOUND
Each loudness prediction took 0.3 ms on one computer processor.
Errors were below 0.5 phons, a difference that would not be detectable. It was tested on sounds, not patients, so it does not change therapy.
Key findings
01A single loudness prediction took 0.3 ms on one CPU.
02After 1,000 training passes, the RMS error was below 0.5 phons for all classes of sounds, a difference that would not be detectable.
03It also generalized well to test sets that were not used for training.
STILL TO COME
How it was doneWhat they found
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 model covered a narrower frequency range than the original loudness model. It was trained with few very quiet real-world sounds, so quiet sounds had larger errors. It was not tested with people who have hearing loss; that extension is only proposed. The paper reports sounds and model outputs, not patient outcomes.
Declared interests
Funded by the Engineering and Physical Sciences Research Council.
The easy way to misread this
Do not treat this as a patient-tested loudness tool. The model was checked against another loudness model for sounds, not against people with hearing loss, and the hearing loss version is only proposed.