Skip to content
HearingTracker
Log in

Predicting whether AI will help an individual with speech in noise

20 replies418 viewsLast reply

Jump
Link to this topic
Reply

I've posted a few times here in the past month about whether I should invest in -enhanced to replace my Costco (Sonova) KS10 HAs, because over the past year or two I've had more difficulty understanding speech in noisy environments (which includes almost EVERY restaurant my wife and I frequent here in Las Vegas.

Because Oticon and Phonak products rise to the top of that heap according to testing here, a few weeks back I visited an Oticon-sponsored dispenser after receiving an email "invitation" to a "45-day free trial." The dispenser performed an and told me I probably didn't NEED Oticon "Reveals" because even my high frequency loss and results were not bad. I left with no new devices but still new questions. Having received another such ad from "Zip Hearing" related to the Phonak Eons that stated I'd be seen by an audiologist, I made an appointment with her. She asked me to bring my latest test scores (which had been done a few months ago at Costco and are posted here as well). My visit with her was VERY low pressure. She told me she didn't "sell" hearing aids herself, but would work with any supplier on a contractual basis and would have her staff make certain I didn't qualify for any device discounts from Traditional Medicare (I'm almost CERTAIN there are none) or from my "Plan F" Medicare Supplement from Anthem Blue Cross (I'm pretty certain there are none available there, either.

But then, she offered some advice I'd not heard before, that makes quite good sense to me. She said that the audiogram basically records frequencies at which the acoustic nerve recognizes sounds, and at what level. What AI chips ad is using a LLM to help the brain translate raw sounds into recognition of what generated them, and that mates with brain processing, likely better for some people than others, and possibly better for one company's LLM than another company's package, and THAT cannot be prospectively inferred from either the raw audiogram OR the WRS results.

I left that session much more satisfied than I had from the Oticon-sponsored dispenser. But, I'm curious whether knowledgeable folk here would agree with her synthesis.

One other aspect of my hearing deficit and its Rx for which I've found a VERY simple management tool is the intrusive amplification of wind noise by my KS-10 HAs when I'm on the road on my bicycle. I just don't wear my HAs, because my low frequency hearing remains quite good, so I'm aware of vehicular traffic approaching from behind me, AND my Garmin Bike Computer, paired with a rear "radar" flasher mounted on my seat post also informs me of vehicles approaching from the rear AND how far away they are as they approach. As the ad for some product I don't remember says, I "don't leave home without (them)."

8 people in this conversation15 min readJump to latest

I think your best bet is to find a good audiologist who has the right credentials and isn't shilling for one brand or another. Have them do their own full hearing eval with you and then ask them what they would recommend for your degree of hearing loss.

There is a lot of hype around anything with the word "" in the name and everyone is trying to cash in on it. The explanation you were given about how it works isn't accurate (there is no LLM involved to my knowledge). What phonak's (and I assume others') systems are doing is using machine learning to extract speech sounds from background noise. Separating speech from background noise is something hearing aids have been trying to do for decades but this method is new. Something to be aware of though is that something like phonak's Sphere works best with a closed fitting. Don't know if you have that or not but that is something the audiologist can explain to you.

@libc said:

What phonak's (and I assume others') systems are doing is using machine learning to extract speech sounds from background noise.

Thanks so much for that correction. My exploration of is typically based on what others have sent me when I ask questions like this. I assumed (incorrectly) that "machine learning" REQUIRED a model to consult in order to make nearly instantaneous adjustments.

The Golden Rule

“All machine learning is , but not all AI is machine learning.”

Just search for “Machine Learning versus AI for hearing aids”.

Thanks for that! Your search suggestion produced a whole BUNCH of advertisements, but also useful information on the meanings of the two terms. I guess that when DO incorporate LLMs, they'll not only customize the sounds that reach our brains, but also TELL us what to eat in that noisy restaurant!

In reply to @jimrobertson

"reads" through large amounts of data that it finds on the internet or in your data and finds information relative to the question you ask. It can synthesize that information into a summary or conclusion about the task it was given. I don't think any hearing aids that claim AI are connecting to the internet or browsing through your personal data files.
AI may have been used to research and create the fixed algorithm programmed into your hearing aid processor but AI is certainly not running in that processor.

may have been used to research and create the fixed algorithm programmed into your hearing aid processor but AI is certainly not running in that processor.

I'm reminded of President Clinton responding to a question about scurrilous behavior with "that depends on what the meaning of the word 'is' is."

User points out that what's proliferating in the hearing aid space now is the addition of or "machine learning" chips, and those can enable AI that uses large language models to create responses to situation such as "what is wind" or "what is useless noise" and more (which many consider an example of "artificial intelligence, but (if we're to believe Apple), a good bit of a large language model's work can occur on device, or (to preserve privacy of the user) at walled garden spaces on Apple servers with the queries or results of queries themselves unavailable to others on the internet. But we're early in the life of AI, and whether it will help us or kill us all is yet to be determined.

@jimrobertson said:

what is wind

Phonak WindBlock and Oticon anti-wind feature are independent from sound processing. However, their activation may be dependent fro scene classifier, which were trained by AI.

I'm a musician and use a few tools that can separate out the vocals and individual instruments from a song. There are also AI apps that can remove background sounds from vocals. They work incredibly well and they aren't even using the latest AI models.

Yes, it can be done. They don't work as well in real time, but they're incredibly fast, so it won't take long before that can be done dependably. Pitch correction for vocalists used to have to be processed after the track was recorded. Pitch correction can be done in real time now. I like it better after the fact so it can be dialed back to sound more natural, but the ability to do it live is still pretty amazing.

Newer tools with and without AI allow me to record and mix music even though I can't hear frequencies over 3-4K. I can visually look for peaks and levels that I can't hear and the software can look for overlapping frequencies from different tracks and fix any collisions that would make the mixed output to be muddy or hard to distinguish separate instruments in the mix. I can still trust my ears for position and balance and adjusting levels and the basic mixing things, but I would need someone else to verify the things that the software allows me to do. I send my finished mix to people to listen to and ask if anything needs to be fixed and they may want a particular instrument to be more or less prominent in places, but that's a preference call as some people like more bass or a particular instrument to be more out front, but that's not a technical issue. Once I get some feedback on the final mix, then AI is used to master the song so it sounds good using earbuds, speakers, car stereo systems, etc... and then it's ready for streaming.

I had been so afraid that I would no longer even be able to do any of this, but the new tools let me still stay in the game. I'm working on a project that should be ready to be released for streaming in a couple months. Getting feedback is the thing that slows things down quite a bit, but I need it before I can do a release. I never thought that any of this would be possible with profound hearing loss. I still have problems with hearing speech, but music sounds fine with the 's That's because music is more important to me than speech. I have a couple programs in my HA's that do work much better for speech, but music and everything else sounds like crap, so I use them when absolutely needed and switch back when I don't. AI should be able to work well for both. I use Widex HA's because their audibility enhancer moves higher frequencies I can't hear down by an octave and everything is still perfectly in tune. That's not a big deal if you only care about speech, but it's essential for music.

AI is more than capable of isolating voices and removing background noise. Ideally, you still want some of the background noise for things to sound authentic, but that just blending a little of the original signal in with the processed one. Real time processing is close if not already there. The issue is more of how much processing can be done through hearing aids or whether it will also need to be linked to a phone or an external server. Keeping HA's as small as they are now and running AI locally is a challenge as AI needs a lot of processing power and memory, so they may need to be a bit larger or in tandem with a phone or with a remote server.

@mike_g said:

Newer tools with and without AI allow me to record and mix music even though I can't hear frequencies over 3-4K.

Fascinating. I'm still new here, even though I'm almost a decade into wearing . Is it the case that HAs cannot make you perceive sounds higher than 4K Hz?

Edited

Yes. With the it's about 3K and with the audibility extender up to about 4K. That's what I can hear as sine waves which are a single frequency. With AE I can barely hear the highest keys on a full piano keyboard with G7-A8 in the 3-4K range. It definitely tops out before 4K. If I can't hear those frequencies at all, then I wouldn't be able to hear them with HAs either. HAs can increase the loudness of frequencies you can hear, but it can't give you hearing you don't have. Audibility Extender reproduces higher frequencies to extend your range a bit but it can only do so much, which for me gets me closer to 4K. I can hear the microwave beep with AE on, but can't hear it without.

I can't understand it exactly, but with music my brain somehow fills out the overtone series or I hear them at such low levels that they're unconsciously perceived but not consciously heard. I guess knowing and remembering what particular instruments sound like may also be part of it. Maybe it's all in my brain, but when I listen to music I can hear sounds higher than I should be able to. I can tell the difference between identical guitars and pianos at higher frequencies that I shouldn't be able to.

Speech is totally different. At normal listening speed I can't always make out all the consonants fast enough to match to the rest of the vocal range that I can hear. So there are some occasional missing words or at least unclear words. Slow down the speech speed and the accuracy improves quite a bit, but you can ask people to repeat something, but asking them to slow down isn't something I'd usually do. I remember when learning Spanish that I could read and speak it at a conversational level, but understanding what people were saying just went by way too fast to catch everything. Common phrases were OK, but the rest was a stream of sounds I was still trying to parse out. Your native language is no different except that you've learned how to parse out words and phrases so naturally that you don't even think about it until you have hearing loss and you do have to actively put things together which sometimes works better than others. And, lyrics are really hard to make out. I miss a lot of words on them.

In reply to @mike_g

@mike_g said:

I can't understand it exactly, but with music my brain somehow fills out the overtone series or I hear them at such low levels that they're unconsciously perceived but not consciously heard. I guess knowing and remembering what particular instruments sound like may also be part of it. Maybe it's all in my brain, but when I listen to music I can hear sounds higher than I should be able to. I can tell the difference between identical guitars and pianos at higher frequencies that I shouldn't be able to.

I have several musicians in my family. If I could plan my own funeral without limits, it would include the bagpipes ringtone of my iPhone (The Skye Boat Song), which QE2 chose to be the last bit of music the world associated hearing with her as her coffin was carried into St. George's Chapel (interesting, because the song came about to celebrate the inability of the British Army to find "Bonnie Prince Charlie" after the highland Scot "Jacobites" were massacred at Culloden in 1765), and at LEAST the last movement of Mahler's 2nd Symphony.

My own relationship with music is so much more pedestrian, yet still abundantly curious. Just HOW could Beethoven compose his 9th Symphony (the absolute APEX of composition at the time he wrote it), while totally unable to HEAR, it other than by a synthesis of MEMORIES. And, how is it, that when I sit in front of my laptop and listen to Mahler's 2nd streamed through 4-generation old Phonaks it makes me ALMOST as happy as sitting in Davies Symphony Hall in SF to listen to it performed live by the SF Symphony under Michael Tilson Thomas's baton?

Could you post an ? I'm 80, but still learning. For example, the bagpipes. There was an article in the NY Times a month or so ago about how HARD it was for pipers in NY City to practice because wherever they tried, the noise ANNOYED at least someone. The pipes can be awful to bear, but also lovely. One thing that helps is tuning the drones accurately. In April 2025 I went to NYC to see my son and watch the "Tartan Day Parade," and I was fascinated to see pipe bands having their 18th Century "drones" being tuned by the world's most prevalent 21st century musical "instrument", as in every band someone would move from piper to piper holding an iPhone running a frequency analyzing App just above the trumpet of each of the 3 drones. When they all produce the same notes, it can be sublime (and even I can tell).

In reply to @jimrobertson

I had KS-10s, and I moved to Infinio Spheres, and I am now trialing Eon Spheres, and I am due to get fitted with Fortell in two weeks.

My low frequency hearing is worse than yours. My #1 hearing challenge has been speech in noise, followed by speech in reverberant or "muddy" acoustic environments (e.g. theater).

I agree with what your audiologist suggested. characterize only part of the phenomenon of hearing loss. Another part deals with localization, or the ability of the brain to differentiate among sound sources based on the location of the source(s), which goes to the ability to detect and process interaural time and amplitude differences among competing sound sources.

Anyway, shortly after I received the Spheres, my two daughters and I attended a performance of Wicked at the Orpheum in San Francisco. At intermission, we were downstairs in the very crowded bar/lounge. Ambient noise level was 92 db (per my Apple watch). I could "easily" carry on a conversation with my daughters without yelling or needing to have someone speak directly into my ear. I was amazed. No way that would be the case with my KS-10s.

Anyway, get fitted for Eons (I also got mine through Zip Hearing) and go out and have a conversation in one of those noisy Las Vegas restaurants. If you don't experience a profound improvement, or enough of an improvement to satisfy your, then just take them back.

Also - I want to repeat the observation that Spheres work best with closed fittings. In my theater experience, I had power domes. I subsequently replaced them with titanium c-Shells, which I really like.

Oh - I am also a cyclist. I haven't used the Eons on my bike yet, but Hear Advisor says the wind performance of the Eons is much improved over the Infinio Spheres.

@garyh said:

Also - I want to repeat the observation that Spheres work best with closed fittings. In my theater experience, I had power domes. I subsequently replaced them with titanium c-Shells, which I really like.

What receiver power do you have? And what diameter of vent?

I am due to get titanium cShell with no vent and UP receiver.

Edited

In reply to @garyh

@garyh said:

shortly after I received the Spheres, my two daughters and I attended a performance of Wicked at the Orpheum in San Francisco. At intermission, we were downstairs in the very crowded bar/lounge. Ambient noise level was 92 db (per my Apple watch). I could "easily" carry on a conversation with my daughters without yelling or needing to have someone speak directly into my ear. I was amazed. No way that would be the case with my KS-10s.

Thanks SO much for that report. Of course, you are impacted by a much more severe hearing limitation than I, but we have some similar experiences and electronics (KS-10s, Apple Watches, entertainment in San Francisco), so I hope you won't mind if I mention a few of those for context.

For example, I purposefully leave my at HOME when I ride, because I have NO trouble hearing cars and trucks approaching from the rear (although, come to think of it, I HAVE once or twice been SURPRISED when passed by another cyclist on a well-maintained road bike—not surprised at being passed, of course; after all, my next B-day will be my 80th). But, occasionally, when I'd go to Davies Symphony Hall I might forget my KS-10s, and while that might compromise my before-concert meal or after-concert dessert, the concert itself would still be thrilling.

When I saw the audiologist a week or so ago, she interviewed me and discussed in general terms what was available as far as HAs endowed with machine-learning chips, and she told me that Zip Hearing's model was using audiologists as the direct patient fitters and adjusters for HAs, but Zip as the device sellers. I'll speak with her again tomorrow, but not in person. In the interim, she will have researched pricing for me. I have traditional Medicare and an excellent "Plan F" Medicare Supplement policy, NOT Medicare "Advantage" (meaning I can see anyone I want rather than being limited to an HMO or PPO plan, but ALSO meaning that I don't have any "coverage" for dental, hearing, or eyeglasses). On the other hand, the value of such "coverage" from "Medicare Advantage" plans is extraordinarily variable. Fo example, my 83 yo brother could be helped by full-mouth dental implants, and his plan "covers" that, but only one a year, so he COULD, theoretically, get them, but he'd be 115 years old when the job finished :cry:

 

Oh, and I just added a post to another thread about the Eon Spheres, because the review indicated they did a bit WORSE after tuning than beforehand, and my own entirely subjective assessment is that I can be enraptured by Mahler live OR recorded with or without assistance from my KS10s.

Related discussions