Skip to content
HearingTracker
Log in

Benefits of ANC and AI speech enhancement for SNR and intelligibility in real hearing-aid recordings

20 replies290 viewsLast reply

Jump
Link to this topic
Reply
Edited

Active noise cancellation (ANC) is widely used in headphones but remains uncommon in hearing aids. Although ANC can reduce occlusion and improve streamed-audio listening in noise, its value for understanding speech in noise is less established.

Because -enhancing algorithms operate only on the hearing aid’s amplified path, the unprocessed direct (vent-transmitted) path limits how much noise reduction reaches the eardrum. ANC attenuates this direct path, allowing such algorithms to reduce noise beyond the limit imposed by a typical hearing-aid fitting.

Our previous simulations of this mechanism predicted substantial intelligibility improvements when ANC and SNR-enhancing processing were combined. Here, we use recordings from commercial devices in realistic acoustic scenes to examine how ANC interacts with an speech-enhancement algorithm.

  • via Andrew Sabin, PhD in his role at Bose Corp. From his IHCON 2026 poster.

Estimating SNR from Recordings

Each recording is split into speech and noise using a binary time-frequency mask derived from reference recordings. Subtracting the noise spectrum from the speech spectrum gives SNR per band.

10 people in this conversation8 min readJump to latest

Wow, very, very interesting.

  1. Not only because of ANC in hearing aids
  2. But also because of using dynamic drivers (custom moving coils) instead of balanced armatures.
  3. It's interesting also, because even with occluding fittings, like tight power domes, there is usually little attenuation below ~500-750 Hz. Example of REOG (Real-Ear Occluded Gain) with power domes here (don't bother with left ear):

So, with ANC it could be possible to have better noise attenuation than with closed fittings 😅🤔 I wonder if it also applies to risk of feedback.

And we don't know about battery draining.

I wonder why an option with multiple balanced armatures (like in the some IEMs) isn't available for users, it's because of current necessity?

I think a key issue is that most people’s hearing loss isn’t severe enough to require completely sealed earplugs, which are necessary for active noise cancellation. You can ask Bose whether their headphones meet this requirement. Open-type active noise cancellation is useless; however, using sealed earplugs solely for this purpose might be overkill.

This was my favorite:

It shows the additional speech intelligibility gains from turning on ANC, where it mostly helps with recovering low-pitch speech information. is required for ANC to work well, but the Bose receiver also gives active occlusion cancellation making it more tolerable to use closed domes.

On current drain, I would think basic physics: more processing, more drain. Wearing totally occlusive cShells with my Spheres, I have noticed low-frequency penetration, but with a cShell, which is relatively massive, the penetration seems considerably reduced from what would reach your ears with an , and perhaps the reduction increases as you move more into speech frequencies.

An interesting set of figures would show the normal frequency spectrum in real speech environments where noise reduction matters, e.g., a cocktail party, restaurant, Costco, or Apple Store. In terms of Spheric noise reduction, my Spheres work very well to clarify speech in almost all situations where I need them. So, I would indirectly conclude that most of the time, low-frequency noise leakage through my very occlusive cShell fit is not much of a problem for me.

With ChatGPT's help, I found the following very interesting paper: https://pubs.asha.org/doi/10.1044/2025_JSLHR-25-000405

It shows that the importance for understanding speech at different frequencies depends upon the listening environment and does not always follow the standard frequency-importance curve used in the poster Abram presented. Compare Figures 2 and 5 in the paper.

The café and food-court conditions placed the greatest importance mostly in the mid-to-high frequencies. The dinner-party condition, however, showed substantial low-frequency importance, including its highest estimate at 432 Hz, along with strong importance around 2.3–3.7 kHz.

I had ChatGPT combine the results of Figures 2 and 5 into a composite figure. However, because posting -generated material is not permitted on the forum, I cannot show that figure here.

I thought Bose abandoned the hearing aid market when they licensed their I.P. to Lexie. It’s encouraging that Bose is continuing their development. Do they or their licensees have any products on the market now?

In reply to @jim_lewis

@jim_lewis said:

It shows that the importance for understanding speech at different frequencies depends upon the listening environment and does not always follow the standard ANSI SII frequency-importance curve used in the poster Abram presented.

Just to supplement the topic about (Speech Intelligibility Index):

https://www.audiologyonline.com/ask-the-experts/speech-intelligibility-index-sii-and-282431

and

https://www.audiologyonline.com/ask-the-experts/what-is-sii-260852

@Bimodal_user said:

Just to supplement the topic about SII (Speech Intelligibility Index):

Thanks, your citations are helpful to understanding . I think something to keep in mind when reading Audiology Online white papers is the experts are affiliated with a company that's selling something, so their analysis may be somewhat slanted to emphasize the approach to understanding or analysis of what their company is able to market. Don't want to sound overly cynical - just think that consideration deserves to be in the back of one's mind, not that peer-reviewed papers eliminate such considerations.

In reply to @jim_lewis

@jim_lewis said:

The café and food-court conditions placed the greatest importance mostly in the mid-to-high frequencies. The dinner-party condition, however, showed substantial low-frequency importance, including its highest estimate at 432 Hz, along with strong importance around 2.3–3.7 kHz.

This is my Achilles' heel (weak point), even with UP receiver I'm missing a few decibels in this band. Look at my UP gain waveform curve, which is interrupted in this band due to receiver limitations (right side):

Fortunately, I am still getting life-changing benefits even with P receiver compared to my old Paradise P90-RT with P receiver.

Edited
@Bimodal_user said:

This is my Achilles' heel (weak point), even with UP receiver I'm missing a few decibels in this band

In the gain view when curves overlap like this it just means you cannot get any more gain but it doesn't tell you about the actual output of the devices. Switch to output view --> real ear on the graph to get a better idea of what is actually going on here. I'd be interested to see the screenshots. Tag me when you post them.

In reply to @Beet

The forum previously mentioned that Orka launched a hearing aid featuring Bose ANC technology in Shanghai this May; a former Orka employee I know mentioned that they began working with Bose in 2024. Did Orka directly use ReSound's M& receiver into this product?

I know this may not sound technical, but sometimes I daydream about custom molds that truly fit MORE of the ear canal, so the sound can get closer to the eardrum.

I have totally cinderblock ears, but find that if I press my tips DEEPER into my ear canal, I get about 8-10% better clarity for speech. Granted, I can't go 'round all day with fingers jammed in my ears, but why can't our ear molds be made longer to go into the ear canal? I have wonderful, comfy roughed-up titanium custom molds that stay PUT alllllll day long (a vast improvement over my previous silicone power domes that I'd have to push in hundreds of times a day), but these are little STUBBY tips that seem to just clear the doorway - like too far from my eardrum to give me optimum speech clarity.

Is there some challenge to getting ear molds made to fit our ears with more surgical precision? Instead, we seem to get the shallow molds that plug the hole but don't carry the sound down the tunnel.

OK. WOW. This is exactly the kind of thing I was daydreaming about - so I'm re-posting the photo of the much deeper-reaching molds right here again.

FUNCTIONAL earmolds by the look of itOOps! I'm new to the new GUI here and may have plunked the image down like ELEPHANT-sized, but even so! This is what I'm talkin' about.

I wonder if I could show this to my audi and she could just TRY to have these made for me in roughed-up titanium?!?!?

I realize that KEY to these custom molds is who makes them. So does anyone know if these deeper molds could be made by anyone out there?

Here is a shot of my stubby Pirates by comparison:

Pirates with custom titanium moldLOL I feel like I'm CLOBBERING everyone's eyeballs with this genormous shot! But I don't know how to make the image smaller for the post here.

Edited

Phonak makes them, but your audiologist has to take a deeper impression of your ear and request them to be made after the first bend until just before the second bend.

Have you worn long earmolds before? They can be a bit uncomfortable and hurt with some movements, especially if they pass the second bend.

I had problems getting the molds right for me for my Jabra 30s. I have a very small ear canal and, according to my fitter, who is experienced with molds, it doesn't have the usual bends. I had a lot of problems with feedback and the aids constantly falling out. With the second set of molds, it was better but not good enough.

The third set of molds seem to be the charm. I asked my fitter to make the molds going deeper into the ear. And she gave them skeleton locks, which prior still didn't keep them in. Now they stay in, and I rarely have feedback. If I do, I just press them in a little more. Going deeper in my ear was the fix I needed.

When I've been wearing them longer, it sometimes hurts my outer ear a little, which, hopefully, can be fixed to have the skeleton lock just a hair loser.

My point is that having them deeper in the ear was critically important.

Related discussions