Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Deep Learning enables hearing aid wearers to pick out a voice in a crowded room

spectrum.ieee.org

Post reply on HN
Print view View original

31–40 of 60 posts

Re: Deep Learning enables hearing aid wearers to pick out a voice in a crowded room

#31

Post by biggieshellz »

chas wrote: post #7

This approach surprised me. Why are they doing feature extraction and then feeding that into a DNN? It seems much more straightforward to have the input of the network be noisy samples and the output be clean samples a la super resolution[0] in images. They probably wouldn't want to use fully-connected layers in that instance, but I don't see any fundamental barriers if they have enough computational power to run a n…

The filter bands they're talking about are Bark bands (https://en.wikipedia.org/wiki/Bark_scale) and are actually representative of the way the ear perceives loudness. In a traditional hearing aid, you might have a compressor for each of these Bark bands to counteract the effects of loudness recruitment (http://www.sens.com/helps/helps_d03.htm).

Re: Deep Learning enables hearing aid wearers to pick out a voice in a crowded room

#32

Post by philipov »

tobtoh wrote: post #6

> The greatest frustration among potential users is that a hearing aid cannot distinguish between, for example, a voice and the sound of a passing car if those sounds occur at the same time. The device cranks up the volume on both, creating an incoherent din. It may be a simplification of the article that I'm misinterpreting, but as someone who got a hearing aid in early 2016, that's not how (modern) hearing aids wor…

I've gone through periods where I didn't wear glasses for 6 months to a year, and every time I would get new glasses, I had exactly that sensation of everything looking too crisp, because I had gotten used to everything looking fuzzy.

Re: Deep Learning enables hearing aid wearers to pick out a voice in a crowded room

#33

Post by biggieshellz »

tobtoh wrote: post #6

> The greatest frustration among potential users is that a hearing aid cannot distinguish between, for example, a voice and the sound of a passing car if those sounds occur at the same time. The device cranks up the volume on both, creating an incoherent din. It may be a simplification of the article that I'm misinterpreting, but as someone who got a hearing aid in early 2016, that's not how (modern) hearing aids wor…

It depends on how the ear was damaged. The cochlea has two types of hair cells: inner and outer. The outer hair cells amplify the incoming sounds by vibrating in tune with them, and the inner hair cells actually pass the signals along to your brain.

If the outer hair cells get damaged, your ear can still perceive all the frequencies; they just aren't amplified enough. In that case, you're correct that boosting the frequencies back up according to your audiogram (with some nonlinear compression to account for loudness recruitment) can bring back your hearing. The same holds if you have a conductive loss (poor transmission of sounds between your eardrum and your cochlea).

On the other hand, if the inner hair cells get damaged, you can no longer hear at the frequencies corresponding with the hair cells that were damaged. The same holds if you damage the connections between the inner hair cells and the nerves, or if you damage the nerves themselves (in which case you may have normal hearing thresholds but still have trouble hearing). In these cases, even if you amplify to match the loss, you can't bring back normal hearing; hence the need for signal processing to make the best of what you have left.

Re: Deep Learning enables hearing aid wearers to pick out a voice in a crowded room

#34

Post by kafkaesq »

Smaug123 wrote: post #11
tobtoh wrote: post #6

> The greatest frustration among potential users is that a hearing aid cannot distinguish between, for example, a voice and the sound of a passing car if those sounds occur at the same time. The device cranks up the volume on both, creating an incoherent din. It may be a simplification of the article that I'm misinterpreting, but as someone who got a hearing aid in early 2016, that's not how (modern) hearing aids wor…

> That blows my mind ... a bit like how do you know the color green is green. Maybe it's purple, but you have been told by someone else that it's green. You may be interested in the "the map is not the territory" idea. "Green" is not a property of an object but of an observer, though "emits light at wavelength N" is a property of an object.

"Green" is not a property of an object but of an observer,

But "Green" happens to be a property of other large families of objects -- especially animate objects (foliage, certain insects, birds, and fish) and, more rarely, certain inanimate but nonetheless "special" objects in nature (features in geology; the sky at certain times; and of course, rainbows).

So in that sense -- while "Green" by itself doesn't seem to have intrinsic properties besides an associate to a certain band of the electromagnetic spectrum -- it does have a strong (extrinsic) association to objects which do have interesting intrinsic properties.

Re: Deep Learning enables hearing aid wearers to pick out a voice in a crowded room

#35

Post by Kluny »

sdrothrock wrote: post #21
Kluny wrote: post #5

Earlier quoted context omitted.

Which one have you got? I know what you mean with the "helpful" bullshit. My conventional hearing aid, on my left ear, has this "smart" mode where it tries to detect speech vs noise, and change the volume or the directionality of the microphone to compensate. You end up with this wildly fluctuating volume all the time where it feels like stationary objects are coming at you. I had them turn that feature off asap. On…

I have the Advanced Bionics Harmony BTE. Since my implant is AB, I wouldn't be able to get the Nucleus Freedom 6. I have an in-ear mic, which does wonders for reducing surrounding noises and also for letting me use a phone normally, but my main issue is with the software itself; I've had issues with it since implantation and they've always been pooh-poohed by audiologists at Hopkins, Tokyo University, and Toranomon.…

Yep, that's exactly the thing I'm talking about. You'd think they could hire one deaf person at their labs to road test the things, but...

On my hearing aid it's a "feature" that can be turned off. Too bad you're stuck with it.

Re: Deep Learning enables hearing aid wearers to pick out a voice in a crowded room

#36

Post by mason240 »

tobtoh wrote: post #6

> The greatest frustration among potential users is that a hearing aid cannot distinguish between, for example, a voice and the sound of a passing car if those sounds occur at the same time. The device cranks up the volume on both, creating an incoherent din. It may be a simplification of the article that I'm misinterpreting, but as someone who got a hearing aid in early 2016, that's not how (modern) hearing aids wor…

>Everything sounded far too crisp.

That's the way felt when I started wearing glasses at age 33. Everything was suddenly in HD.

Re: Deep Learning enables hearing aid wearers to pick out a voice in a crowded room

#38

Post by dhuntermn »

stefs wrote: post #18
tobtoh wrote: post #6

> The greatest frustration among potential users is that a hearing aid cannot distinguish between, for example, a voice and the sound of a passing car if those sounds occur at the same time. The device cranks up the volume on both, creating an incoherent din. It may be a simplification of the article that I'm misinterpreting, but as someone who got a hearing aid in early 2016, that's not how (modern) hearing aids wor…

i'm 35 and i've got HAs for 20 years now. my hearing loss is quite severe (without aides i don't understand face to face conversation without the other person being close to shouting). modern devices employ another trick: they "compress" frequencies you're hearing less into a spectrum that's not as damaged (this is mostly done if your hearing loss affects the frequencies used for speech). the side effect is that musi…

I got a new pair of Starkey HAs recently that have a separate processor for music. The music mode turns off compression completely. Even in normal mode, it appears to me that the music processor is being used, because music sounds so much better. I recently had a pair of Altec speakers that I've owned since the early 70s rebuilt, and was listening to Pandora via Chromecast Audio. The song Spill the Wine by War came on, and I could actually hear the strings of the bass guitar vibrate. Before I had these new HAs, bass guitar just sounded like vague, low notes in the background. I was amazed and overjoyed.

Re: Deep Learning enables hearing aid wearers to pick out a voice in a crowded room

#39

Post by nabla9 »

jfsantos wrote: post #37
nabla9 wrote: post #30

How is this method better compared to independent component analysis?

For one thing, independent component analysis needs to process signals from as many microphones as there are sources to work properly.

This is true for vanilla ICA.

Independent Subspace ICA models can be applied to more signals than sources problems. It's also possible to use different decomposition methods, or subtracting already detected signals.

Re: Deep Learning enables hearing aid wearers to pick out a voice in a crowded room

#40

Post by adinisom »

chas wrote: post #7

This approach surprised me. Why are they doing feature extraction and then feeding that into a DNN? It seems much more straightforward to have the input of the network be noisy samples and the output be clean samples a la super resolution[0] in images. They probably wouldn't want to use fully-connected layers in that instance, but I don't see any fundamental barriers if they have enough computational power to run a n…

That might work, although I think there are two limitations:

1) Hearing aids have a 10ms latency budget. So no matter how much processing they can do, they're limited by how many samples they can look ahead and that limits the design of the filters. The brain can presumably look ahead further to separate sound streams so I think it's pretty impressive that ideal binary masking works.

2) Hearing aids have a power budget. The ones I've looked at achieve low power by running a FIR filter in hardware to shape the sound while a DSP classifies the sound and adjust the filter taps. The DSP doesn't have to run at the same rate as the filter. That seems well matched to the binary filter approach. Likewise features extraction might not run at the same rate as the DNN.

Post reply on HN