Benefits of AI and Neural Networks in hearing aids

Not an audiologist or a manufacturer, but I am a software developer, for whatever that’s worth.

The basic idea behind AI in hearing aids is that good, low-effort speech comprehension in noisy, acoustically complex environments requires more than just amplifying frequency X by Y decibels based on your audiogram. At their simplest - which isn’t really all that simple - AI techniques can be used to “train” the main operating program of a hearing aid (like Phonak’s AutoSense) to recognize different sound situations and apply sets of pre-configured settings and algorithms (including amplification, volume compression, frequency compression, noise suppression, and so on) to give you good results as you move from quiet to noisy places, drive, listen to music, or do whatever else strikes your fancy. AutoSense, IIRC, is even supposed to be able to blend two different programs when you’re in an in-between situation. The result, at best, is that you don’t notice all that’s going on under the hood - you just go through your day and hear what you should be hearing without thinking about it.

The latest application of AI in hearing aids goes quite a bit beyond that: The Phonak Spheres (and, I believe, at least one or two competitors) have a real-time special AI processor (a separate chip in the Spheres, thus their large-ish size) that uses a number of inputs, including motion sensors and directional microphones, to continually evaluate exactly which sounds coming from which directions are likely to be people saying things that you’re interested in hearing. Those sounds are amplified, and the remaining sounds are rather aggressively suppressed, with the result being a vastly better experience understanding conversations in noisy environments. (This works best with closed domes, power domes, or custom shells; for an open-dome guy like me, a fair bit of background noise will get through with no help or hindrance from the HAs. But most open-dome people probably don’t need the extra help as much anyway.) Because this “Spheric Speech in Loud Noise“ mode is designed to cope with situations that are otherwise very difficult for people with hearing loss, it’s not as subtle and natural-sounding as the other, “canned” automatic programs - as a general rule, Spheroids seem to notice a distinct change when the AI kicks in.

While I have the Spheres, I can’t personally vouch for their noisy-environment superpowers - I’ve hated noisy environments my whole life, even with the perfect hearing I had till recently, so I’m very seldom in places where “Spheric Speech in Hell” would be useful. (When I am in such places, I’m usually struggling to get out of them as fast as possible.) But at least a couple of people on the forum have experienced remarkable results with Spheres - they really are pretty amazing little (?) gadgets.

What all this means to you depends on your hearing loss, your lifestyle, your financial situation, and so on. For me, the Spheres were my best choice because of their excellent, flexible Bluetooth Classic capabilities - I actually might have chosen the non-Sphere Infinios, which lack the Spheres’ real-time AI chops, had they been available here. If you like to socialize in restaurants and coffee shops (or construction sites or bowling allies, I guess) but you have trouble carrying on conversations in these places, you may find a properly configured set of Spheres (or equivalent) to be extremely helpful.

@tenkan said:

I didn’t bother reading past this point, I don’t believe AI in HAs can be trained/learn on the fly, it’s all data points built in.

Apparently I was unclear - in that paragraph I was referring to training AutoSense before its release, not to real-time learning when the Spheres are being worn.

In fact, all Phonak’s references to “training” are in the context of their development process; to my knowledge, none of the hearing aid AIs to date actually learn on the job.

Eventually, of course, we will have “reward” and “punish” buttons to help our HAs improve. I can hear their tiny screams of agony already… :-o