Skip to content
HearingTracker
Log in

AI in Hearing Aids

35 replies1.7k viewsLast reply

Jump
Link to this topic
Reply

Anyone on this forum technical enough to speak about these claims the hearing aid manufactures are making?? I’ve worked in IT for over 20 years but I’m not a programmer, I’m an analyst. Generative AI takes way more memory and processing power than a laptop, phone or any type of personal computer has available. For work I have a Microsoft Copilot license, if I disconnect from the Internet, it won’t work. It requires large scale data centers.

A press release from Starkey mentioned using neural network for detecting speech in noise which makes sense. This sounds like an excellent use of AI. But I’m still skeptical given the processing required and a hearing aids limited battery life. Plus, there are lots of stories about how AI got the “wrong answer”. AI works on probabilities, so there are going to be a small percentage of the time when AI won’t make the right adjustment and the user won’t hear anything.

I’m wondering if these manufacturers are actually using a sophisticated algorithm and calling it AI is just marketing hype. Does anyone have any additional, in depth details to help me understand how this is more than hype?

Thanks,

Jeff

21 people in this conversation17 min readJump to latest
Edited

I use the Starkey Edge . If it’s real AI I don’t know. I do know it works great when I use it in noise.

I’ve been to restaurants that are crazy loud. I have no problem hearing and talking.

@jmwall67 said:

I’m wondering if these manufacturers are actually using a sophisticated algorithm and calling it AI is just marketing hype.

Yep you can’t believe everything you read from the marketing department blurb, it’s mostly “hype” as every major manufacturer is making the bold statement of “hear” everything in every situation..but the future is looking pretty good, it’s early day’s now.

Well it’s been reported that the hearing aid uses the battery more quickly when a hearing-in-noise mode is selected, and it was one of the reasons for going with a rechargeable battery.

I think there are people on the forum capable of understanding what goes on BUT the manufacturers are very secretive about how things work so all we’re left with is opinions. To the best of my knowledge, Sonova (Phonak) is the only manufacturer who uses a separate processor for the neural network. Depending on who’s advertising you read it’s either a plus or a minus. The downsides of that extra CPU are increased hearing aid size, mainly to support an adequate sized battery to provide decent battery life.

Phonak has a couple of white papers on their processor on their website. Yes, I worked in process simulation software for thirty years. Yes, strictly speaking, this is not .

Phonak’s processor is trained on millions of human speech sounds and uses a neural network to do millions of comparisons per second of the sampled speech in realtime, then puts out the best one. I think the other manufacturers do the same.. it’s a neural network that makes it work.

While Phonak uses a separate chip, other manufacturers like Otocon and Sharkey have theirs integrated. Phonak’s is a battery hog, and I believe the next generation Phonak will be integrated also.

I chose Phonak simply because the other competitors didn’t support the my Android phone all that well. That was a year ago.

I think the mfrs generally program a programmable array to make control inputs to the signal processors based on conditions, and the array is programmed using a large amount of sound samples which have been characterized as “good” or bad. This is a very simplistic description. Google “use of fpgas to control dsps” to see maybe more than you want. Think of an fpga as a huge network of circuits that can implement a sophisticated “if-then-this-except” condition/response control function. It has programmed decisions burned into it, and they don’t change unless you get a new firmware update from the mfr. no “learning” as it goes, it learned everything when it was programmed in the lab before the firmware was loaded.

WH

I’ve been using the Phonak Sphere’s for a year and I can personally confirm that they are game changing in noise. So much so that I can often hear better than people with normal hearing. That being said, it does appear that they work better for people with closed fittings and flatter types of hearing loss. Maybe that is why they work so well for me.

I also want to say that the use of in hearing aids is a very, very simple and purpose built AI application. It’s sorta only doing one thing all the time and that is removing noise.

Jordan

Thanks for all of the thoughtful replies. It makes sense to me that mfg’s would train a neural network and then load the results into the hearing aid.

This was helpful in my understanding of how the new technology is being implemented in hearing aids.

Jeff

In reply to @CaptainTrips

Will be interesting to see what approach wins out regarding separate chip vs integrated. Certainly in gaming computers a discrete GPU provides better performance but integrated GPUs are good enough for non gaming tasks. I think it will come down to whether an integrated neural network chip can be “good enough.”

Edited

I agree. The one good thing about the separate chip approach that Phonak uses is that you get crazy battery life when you don’t engage the chip. I can go two days on a single charge if I don’t use the AI mode in a noisy place.

These AI hearing aids are only going to get better as this AI tech continues to be refined.

Jordan

Something rather counterintuitive to remember is that it’s actually easier to for hearing aids to do well with speech in noise for somebody with a moderate to severe loss. It’s going to take some noise reduction ability we don’t yet have to greatly reduce noise for people with good low frequency hearing. For them the noise is quite prominent without hearing aids. For somebody with a severe loss, the noise does not exist unaided.

I have had Oticon hearing aids for the past 8 years. 2 years ago switched to the Oticon Real model and there was slight improvement but still a big challenge in noisy surroundings. Based on the great results people have expressed in this forum, I went to audiologist and told them I am interested in the Phonak Sphere.

They first had me try the Starkey aids, but the receivers were too long and left ear had an issue with fit. Then moved to the Phonak … I am week 3 of trying them and hate vs. the oticon. I do believe that the programming may be some of the reason, but I cannot hear as well at any time vs. the oticons.

Comments in this forum have stated that people with severe loss will gain the most and would also help. I have moderate lost in the high ranges and still use vented domes so that could be part of my experience.

Seeing the doc on Monday to see what they say about them, but at this point I am ready to go back to the Oticon Reals and wait for a couple more years to let things progress a bit more.

Brian,

I have a somewhat similar loss and have Sphere i90’s. The sphere’s work fine for me and “I BELIEVE” they are the best aids I have had in over 25 years of hearing aid bliss…

I also have Phonak slim tip AOV vent ear-molds. My low frequency loss is minimal therefore Phonak uses my prescription to create a vent in the ear molds sized with their software to allow some of my LF natural sounds to reach my eardrums but also not allow my hearing to be overwhelmed by ambient sounds such as you might receive from standard open domes.

Anytime you enter a noisy environment with open domes and Spheric mode kicks in you are still hearing the ambient sounds mixed in with the hearing aided sounds.

Before kicking the Sphere’s to the curb, may I suggest trying a vented dome, or something vented with a bit more occlusion, or go all the way and try slim tip ear-molds.

I had Oticon Mores prior to getting the Spheres and the difference in sound took time to adjust to but I was amazed how the Spheres allowed me to hear in very loud situations - something the Mores didn’t handle well for me.

I also wish I had tried the Oticon Intents prior to the Spheres since I’ve heard others rave about their excellent speech in noise capabilities.

Good luck with your Monday appointment.

David

It is . 5 of the 6 big brands have . (Widex only uses machine learning) Some are trained on millions of sounds and combine that with environment, sensors, app user input, etc to figure what to enhance and what to attenuate. has gone from around 6 only a few years ago to 12 dB. Others rely more on user preferences or crowd sourcing to make decisions. Oticon and Phonak are mostly automatic while Starkey (at least til now) is more user activated like Signia who thinks the user should turn on AI when needed “like an e-bike”. They are all different in their approach.

Since there is no objective definition of “”, anyone can make any claim about how they are using it. In general, AI is software that can examine and evaluate a very large quantity of data and return an answer based on the query. Using AI for something is like saying “I use a spreadsheet for xxxxx”. xxxxx could be a complicated fluid dynamics problem or balancing a checkbook with $10 in it. My point is, AI is today’s buzz word which is being thrown around as if it is magic that turns a sow’s ear into a silk stocking, or maybe it can’t.

@jmwall67 said:

I’ve worked in IT for over 20 years but I’m not a programmer, I’m an analyst. Generative AI takes way more memory and processing power than a laptop, phone or any type of personal computer has available. For work I have a Microsoft Copilot license, if I disconnect from the Internet, it won’t work. It requires large scale data centers.

I recommend read the Phonak White papers from this site:

Evidence Library | Phonak8

Phonak puts evidence-based research at the core of every new product and feature. Browse these resources to learn more about the science behind our technologies.

Especially from 2024, like:

Brian,

To add to what others said. I have used my phonak Spheres with open domes and titanium slim tips on active vents. The active vents with sealed slim tips is a huge change in my noisy environment listening, for the better.

Related discussions