Has anyone tried the new Whisper subscription hearing system?

I’ve looked into their web now (didn’t occurred to me earlier), so they have the same commonly sized bte units plus pocket unit. That also means 2x wireless transmission latency. Over air is always always slower than in the circuits.

We usually are aware of the one transmission when we stream (but we also can pause the video or something), but regular hearing doesn’t have it. This device introduces latency, double, for regular hearing. That will be awkward. Especially for lip reading people.

Not to mention risk of losing the unit if it’s in the shirt pocket.
And everlasting potential issues with connecting to the ear units - BT and similar wireless protocols need walls to bounce off of, because they cannot go through our bodies. So out in the open, in best case this means holding the unit in front of your face so that sound can be transferred to both ear devices. Worst case is that even that won’t be able to send sound to both aid.

For regular hearing purposes?!
Makes no sense.

If you’re sitting in closed spaces (where transmition should work nicely, unless there’s interference with other devices), then you probably can save a bunch of money by going lower tech level anyway, and you will not have little brick in your pocket.

The more complex listening situation, so that there actually is need for ‘extra processing power’, usually is combined with more complex environmental situation considering wireless connectivity.

Basic purpose of hearing aid is to help us when we talk with someone face to face.
Phone, BT, external mics are additional perks that could help us but come with a potential issues. We’re usually aware of them or learn to cope with them along the way.

Introducing such issues into regular hearing aid need, makes no sense at all. Especially since there is decades of aids that do that basic job without any hassle at all.

Not to mention they’re implying that theirs is better solution…

Other things they use to promote is generic babble, updates are common with big six anyway. They didn’t say anything special that could piqued interest that they have something worth exploring.

They have zero information about setting it up, so I assume only proprietary algorithm, which probably doesn’t allow for following best practices, not to mention rem based live speech mapping fitting.

They didn’t even boost how they provide natural sound, speech in noise and the usual mumbo jumbo. Amateurs :joy:

And they’re asking for the prices as those big six. Lol

I’d probably recommend that only to someone I really really hate :joy: really

Others - run away.

4 Likes

I don’t think you’d call it a subscription system. It’s a three year lease for 5000 USD (at the introductory price). It may not be much different from principle from some of the apps that are starting to appear on smartphones. It will be interesting to see the first reviews (won’t be me).

3 Likes

I think it’s an intriguing concept, and could perhaps even leverage the processing power in a phone (rather than a dedicated box), HOWEVER the unit has to get the audio, process it, and return it to the earpiece. That means continuous battery usage, unless there was just some option to enable it when in a difficult situation.

1 Like

Thanks for the info so far. I was curious about this.

If the technology isn’t a giant leap in helping me, I don’t see it worth the price. Perhaps for others it might be right.

Bob

I ran across this article

A company called Whisper is offering subscription based hearing aids.

Ever hear of them? What do you think?

The Whisper site itself is here:

https://whisper.ai/how-whisper-works/

Personally I think they want too much per month. At least for my budget.

So I’m just curious as to what you all think about this.

Bob

Hi Bob. I think this brings up a big question: If you could get a HA that was an absolute quantum leap better than the best
HA that was available now, how much would you be willing to pay for it?

$139 per month x 36 months is $5004. $179 per month (their non-discounted rate) x 36 months is $6444. That is a little more than I’m paying for my Oticon Mores.

From the web site’s info it seems that you would be able to keep the hardware after 36 months and continue to receive updates. It would be like a Tesla car: They would just keep on sending you software updates no matter how old your car was. If so, that could theoretically be cost-effective compared to a conventional HA.

BUT…I think that sounds too good to be true. I’ll bet you would have to upgrade/replace the hardware every three years or so. Just like with conventional HAs, the manufacturers could probably deliver software updates to current HAs. But they, of course, would rather require you to buy new HAs because they make much more profit that way.

SO…the big question remains: How much better, if at all, do these Whispers sound compared to conventional HAs? I am frustrated that, although the Whispers have been available for a few months now, I can’t find any comparative reviews.

I tend to keep my aids longer than 3 years, usually 5. I don’t have a lot of money, so I stretch things as much as I can.

If they went down to $100/month I’d consider it.

Bob

There’s only 1 YouTube review I found about it (see link below). And it’s not even a user review, more like an affiliated HCP authorized to dispense it trying to advertise it.

The idea of having to carry around that brick with you everywhere you go is really not enticing to me. It’s not worth the hassle at all in my opinion. Even a small intermediate streamer worn around the neck like the ConnectClip is already considered a hassle to many, let alone something that big like the Whisper brain. It becomes more like a ball and chain, not a brain.

Major HA mfgs are able to fit AI circuitries inside their HAs already. For Whisper to have to use a brick to achieve the same thing shows that they’re not heavyweight enough to be worthy contenders.

Their differentiations, from having a separate “brain”, to a rather expensive monthly subscription model, don’t really appeal to me at all. Maybe that’s why there are not enough takers to generate enough reviews. Maybe someone should write to Dr. Cliff to ask him to do a review on it.

2 Likes

I attended the webinar today that was mentioned at the end of the video that Volusiano cited. It was probably very informative for someone who had no idea about these Whispers. For someone with detailed and perhaps challenging questions like I had, it was not that helpful. There were maybe 50 people in attendance and questions were only allowed by chat (text). They said they were going to record the webinar and make the recording available online.

The main thing that I learned was that the Whisper brain has maybe a hundred times more processing capability than the most recent top-of-the-line HA. Does that processing capability result RIGHT NOW in significantly better speech recognition in noise? I asked that question but they did not answer it. To be fair, the webinar was only one hour long and I’m sure there were many questions.

My gut feeling is that this is the future of hearing aids. I, for one, would gladly carry a 4-oz. “brain” in my pocket if it would result in a DRAMATIC increase in speech comprehension in noise.

As always, the devil is in the details. I visit the Seattle area (where the audi sponsoring the webinar practices) fairly frequently. If they would allow me a one-week trial as they said in the webinar, I will try them and post a review comparing them with my new Oticon Mores.

3 Likes

No, but I’d lug it around to challenging hearing environments, 'cause as it is I get nada from being in a room fully of people. But at 6 or 7 thousand (AUD) for a part-time hearing device it’s not going to happen.

That’s a pretty fundamental question not to answer! We’d all love to hear your thoughts if you get to trial it.

2 Likes

Yes @ziploc, if you can do a trial on it, and especially with you owning a pair of the More 1, which is supposed to be driven by the AI DNN modelling, I think you’d be in a unique position to be able to judge if the Whisper with its much more heavy duty processing power will be able to blow the More out of the water or not. I’d be very interested to hear what you think.

My gut feeling is that similar to the initial finding that the More, despite its newly touted DNN approach, is found to be only an incremental improvement on the OPN, the improvement of the Whisper brain is not going to be dramatic enough to justify people being willing to carry around the brain with them at all times.

Remember that the More has a built-in deep neural network that’s already a finished result of AI self-learning to the tune of 12 million sound scenes collected over time and using heavy duty processing-power and tons of equipments with extensive memories over years to develop. However, all these external processing power and stuff used for the training don’t need to be squeezed into a chip small enough to fit into a hearing aid. The extra hardware and processing power is just needed up front externally to do the training, then only the culminated result of the finalized deep neural network “brain” already formed with the appropriate weights and biases is needed to go into the tiny hearing aids to do the final processing, which is a lot easier to implement in a chip. A rough analogy of this is Einstein’s special theory of relativity culminating in E=MC2 (squared). You don’t need to know much else, you just need Einstein to arrive at E=MC2 for you, then you use that simple formula to calculate everything else.

It would be nice to understand what all the processing Whisper needs to do that requires that kind of processing power is. Perhaps it’s better to take the approach Oticon did, use external processing power to get to the final result, then implement that final result into the HA itself.

I don’t know whether to laugh or scream, the peddlers of snake oil is what comes to mind here, these guys are way out of their respective depth, looking at the website makes me think a 3rd grader designed it, it’s completely lacking in any detail about the 'brain" in fact there’s no technical or data details or support on any of the claims they make, no white papers or peer review at all that I can find… It’s one of the first things you push when you truly know you have the “brain” that everyone’s gonna want… right?
You have to charge the “brain” overnight plus you got 675 batteries in the "earpieces… freaking hell it’s like frankenstein all over again, I mean where does that brick of a “brain” reside while using, like in your trouser pocket or around your neck?or maybe just place it on a table in a group setting?
There’s absolutely nothing that this “brain” can do that any modern HA doesn’t already do, save your time and money and get something that has been proven to help (and probably a lot less cost)

1 Like

Obviously a hearing aid pro or at least an experienced HA wearer is going to have to try the Whisper and post a review. Hopefully that will happen soon.

1 Like

+1
I remember my grandfather carrying a “brain” hanging around his neck, connected to a small speaker in his ear.
Circa 1948.

That depends on what snr you need and what is your current wrs.

In short, if you are below 80% wrs and need more than 5db snr, only tech that can help you is skipping the cochlea which is damaged and messes up the signal.
And any additional noise (=unwanted sound) makes it significantly worse for comprehension.

Maaaaybe if you’d be in completely isolated blob, like a space suit look so that you don’t get any noise from the outside, and mic on the person’s mouth and heavy noise cancelling in between, you might come up close to that wrs in quiet.

No such tech exists yet.

Anything that relies on your ear drum and cochlea is limited by your cochlea damage.

However, tech in current implants isn’t still there to give you 100% wrs and tech still struggle with noise as well, because filtering unwanted sounds from wanted is hard, especially when unwanted sound is of speech type (speech babble, bg music, tv and such).

Maybe when idea behind oticon more and deep neural network comes for implants we might see the much wanted kick implants need.

Or maybe they invent real cochlea they could implant, and not just electrode array with limited regions.

The best you can do for now is use aids that allow streaming, and do focused trainings for increasing comprehension out of the mess that comes through the aid. Also, use external mics to help you catch the signal closer to its source.

So, cheapest and kick ass solution would be roger select iN version and phonak paradise 30. Or marvel 30.
P90 alone cannot do much if you’re below 80% wrs and need more than 5db snr anyway, so if you don’t need/want other bells and whistles, investing in excellent mic is better way than investing in most expensive aids.

Yeah, I do have 90, and they do better than 70, (so I didn’t even try 30/50 which have a mic less), but if situation is complex, I can shut down and nod, or pull out the mic and struggle and catch something and follow the most of the conversation unless the noise is really hard, then I’m just doomed. But even in quiet, what mic is able to deliver is miles away from aids alone. Like, I can understand people without looking at their mouth…
My wrs is 70 and snr is estimated at over 15. Doomed :rofl:
But I intentionally trained (in periods), and I can safely say that some positive difference is seen now as compared to almost 9 months ago. Not on the graphs though.

Or another solution which might be similarly awesome and cheaper is starkey, they have table mic, and you just pick the cheapest aids that work with it. But that one I haven’t tried personally so I cannot vouch for how exactly it works.

Or this

Blacky, my bad wrs is from an acoustic neuroma so I don’t think a CI would help. Of course you’re right: even a drastically better AI-based HA is not going to help speech recognition in my bad ear that much.

In fact that was illustrated today. I got my second vaccine shot and while I waited for 15 minutes afterward to make sure of no bad reaction I was seated where most of the ambient speech noise in the room was coming from my left side (into my neuroma-damaged ear). The voices actually sounded louder and brighter with my new Oticon Mores than I have been used to, but they sounded distorted. I guess I have to lower my expectations.

Antique techniques like FM or high-rate PWM give “zero” delay (<1uS). Even 2-way is inaudible. They may need to be “kept close”; pocket, not table.

The technology to put a nano-super-computer on a 1.4V button battery is a brutal investment. Putting an ARM in a pocket is much easier for a start-up. The physical system, and also the programming. I think it could be good to have another player with different thinking. I don’t think I will put my money on this.

$140/month does not seem SO high? Take a pair of HAs at $6000, use for 5 years, 60 months, is $83/month; and while upgrades are possible they are limited by tiny-tiny CPU. This new thing may have a more powerful or more general CPU; it is possible a new pocket-unit with 4X the processing power (Moores Law) could be sold for the price of a rPi ($80) plus professional markup (so $500). OTOH they could get 99 customers and go broke.

I read some of the press releases on the Whisper system and its implementation sounds very similar to the Oticon More approach using the deep neural network learning to refine and improve on the balance of the sound scene.

But the difference is that Whisper uses the “brain” processing power to continue to learn and train and refine its deep neural network (using the “pre-trained” in-house data as their starting point) as usage continues by the users. The Oticon More, on the other hand, pre-trains its neural network with its in-house data (of 12 million sound scenes) up front, then implements the final result of their neural network into the More.

So the Oticon big “brain” doesn’t have to be carried around like the Whisper “brain”. It was already all trained up front and only the much more simple-to-implement final trained neural network can go into a much smaller form-factor inside the HA behind your ear. Meanwhile, Oticon can continue to refine its neural network further externally, maybe with another 12 millions sound scenes in-house, then simply do a new firmware update to plug-in a newly more refined neural network into the HA later on.

I think this is a much better approach than the Whisper approach. It eliminates the need to squeeze the self-learning “brain” into a little box that users are forced to carry around with them every day. Do also note that Oticon trains their neural network using a special 360 degree array of microphones device for complete 3D capture of the sound scenes. Meanwhile, the Whisper system only has maybe 4 mics, 2 on each of their HAs, to capture new sound scenes from users who wear them to further train and refine their neural network inside the brain.

Also, each Whisper neural network inside each individual brain becomes trained in a customized way that is unique to only that particular user’s listening environment. It doesn’t get to benefit from other users’ own listening environments in their own unique brains, unless Whisper somehow has a way to collect and upload all the customized neural networks then combine them and download as a firmware update back to the individual brains.

At any rate, at some point, there’s a diminishing return once the neural network is trained sufficiently well enough that any further training will only very ever slightly marginally improve its performance.

Aside from all this technical stuff, the Achilles’ Heel of the Whisper system that will lead into its early demise will be its brain. Nobody is going to want to pay $140/month to wear something in their pocket with them all day long, when they can get similar performance from something much more portable behind their ears like the Oticon More. Unless the Whisper system can give a much bigger wow factor than the More can deliver, which I really doubt it can, then it’s already doomed from the get-go due to its brain.

What Whisper ought to do is to start its subscription at $20/month at least in the first year or so to encourage people to try it out to get them hooked to it on its own merit. Then change it to $140/month after the first year, once it’s been able to demonstrate its wow factor and its worth to the users. That’s the only way a tiny start up like David, I mean Whisper, can take on the giant Goliath like Oticon and win.

If my assumption is wrong about Whisper continuously training their deep neural network on the fly in their “brain” unit, and they only pre-trained their neural network then implementing it into the brain unit like Oticon does, then they’re already beat by Oticon because Oticon was able to fit it all into a tiny hearing aid behind the ear and requires no external brain unit.

Afaik, you train neural network with one set of inputs and then you verify how it works with another set of inputs.

Because you want to be sure results are what you expect / want, and not something else which lies in the data and network concluded that, but isn’t useful to you.

That’s why it’s important that training set should be as varied and unbiased as possible. Which is hard to do in controlled environment.

Giving neural network to grow in uncontrolled environment could lead to very unwanted results.

Rough example, if you feed it for picture recognition of animals, and you always gave black dogs and white cats, when you show the picture of black cat, it will probably sort it out as a dog.
No clue how complicated consequences in hearing usage could be, but for example, it could learn that all beeps are unwanted or something along those lines.

Neural networks are great, but if you want them to perform in some way (I suspect they/oticon used supervised learning for this one, because that makes sense for the use case - you have outcome in mind and you’re not doing explorations alone), you have to know what you’re doing exactly.

Unless whisper implements also a/b testing and getting feedback from user, as a smallest viable verification measure, then I don’t see how such network could be reliable…

As an experiment to learn from, yes, definitely.
But as a ready made solution, expensive one, for customers? No.