New DNN principles in hearing aids

Most start ups disappear, very few ever really go to production.

Maybe it was a prank? But Giles Tongue is real man Giles Tongue - Bango | LinkedIn10

Creating a junk email address for such uses can be handy.

Yeah, I thought about doing that. But this is the first and only time I’ve run into this so far, and if it were something else I really wanted, I would have done it already. But this whitepaper is not that important enough to me for the trouble anyway. Plus I object to their email requirement in principle in the first place.

@Volusiano:,When are you going to break down and buy some More1s? They’ll rebalance your 🦤 seagulls 🦤 big time!

I’m quoting myself here just to offer a paraphrase of what Oticon says in their literature, based on my own personal experience.

@SpudGunner said:

Given that Oticon’s DNN approach is basically a “black box” to anyone not having access to Oticon’s proprietary data and guided learning protocols, I wonder how useful it is to spend a lot of time hypothesizing about how, exactly, it works.

My mind is happy with this paradigm:

Algorithmic approach: - engineers identify classes of noise that interfere with speech comprehension and write algorithms to recognize and suppress them.

DNN Approach: - engineers teach a black box to identify speech as it occurs in real time and train their noise suppression algorithms not to impinge upon the speech trace.

The way my More1s operate is quite inscrutable to me, but the result is clear: I can hear more sounds around me than I’ve been able to hear in decades, and I can understand the speech presented concurrently to me more easily than with any of the othe devices I’ve worn.

At the beginning, wearing More1s was confusing, since I had gone 6 years without wearing hearing aids regularly, and my brain hearing ability suffered.

As I persisted with my acclimatization, the panoply of sounds gradually began to sort itself out and individual sounds fell into place and made sense.

As @cvkemp often notes in his posts, persistence is the key to success. The longer you wear More1s, and the more correctly they are fitted, the more your brain will gradually regain its ability to decode and understand the speech patterns presented to it.

This is, by definition, an incremental process: there was no WOW factor for me. My brain had lost its ability to understand speech, and it’s taking a while to get it back.

I feel I’m still detecting improvements in my hearing on a weekly basis, and - although I don’t understand how it works - the DNN that resides like a genie within my HAs continues to work its subtle magic for me as conditions demand it.

That’s plenty good enough for me!

@Volusiano said:

Either that or if I can find a pair of More 1 on eBay for a good deal. Looks like you can get a brand new pair of More 1 on eBay from South Korea for around $2800 now. Hopefully the longer I can wait, the lower goes the price.

Hope you get yours soon. Waiting in line too.

@gkumar: How come they’re so much less expensive coming out of South Korea? Are they “grey market”?

Thanks to @flashb1024, I was able to easily get a fake email address to download the Chatable whitepaper to review.

They claim that DNN is used 2 ways: 1) Inline DNN and 2) Hybrid DNN.

Inline DNN means that the DNN is a full-blown DNN inside the chip that actively analyzes the sound then processes the sound then delivers the sound to the user. They claim that this method requires a lot of processing power that introduces up to 40+ ms algorithmic latency between the input sound and the output sound, which is not acceptable to the listener. Chatable claims that they’re the only one who is able to break this algorithmic latency barrier to reduce it down to below the threshold of detection latency of 6 ms. In fact, they claim that their latest AI v3.0 Edge has only 2.5 ms algorithmic latency. Their AI v1.0 was 35 ms and their AI V2.0 was 15 ms. They claim that typical industry AI for speech enhancement is 40+ ms.

The hybrid DNN approach uses a different (more “lame?”) DNN that “passively?” conducts sound analysis (I don’t really know what they mean by passively anyway), and instead of actively process and deliver the sound, it uses the hybrid DNN outputs to adjust the DSP control parameters like how a traditional HA would do, thereby allowing it to avoid the latency issue. They use the analogy of having an audi inside the HA rapidly readjusting the DSP parameters to suit the changing environments.

Well, my take is that the Oticon More has the inline DNN inside its chip and I’ve never heard of any More user complaining about hearing noticeable latency. So I don’t really know on what basis Chatable claims that they’re the only one who’s been able to break the 6 ms latency barrier.

It’s also my take that the Hybrid DNN approach as described by Chatable sounds a lot like the autosensing of environments done by many other major HA brands whereby they use the autosense result to determine what kind of environment you’re in and automatically switch to the appropriate program with DSP parameters suitable to that environment.

So I guess my question to the More users out there on this forum is whether you guys notice any kind of audio latency in the More that would bother you? I don’t really know what the number is for the More, but it doesn’t matter what the number is as long as it’s not noticeable. Of course hearing things without seeing what you hear at the same time does not introduce a latency issue, even if a large latency is there. But if you look at somebody speaking directly to you and their voice seem to trail behind their mouths, then yes, that’s a latency issue.

Another possibility to consider is that while Chatable has been struggling with their blinders on trying to grind down the algorithmic latency in their DNN, maybe Oticon has found a different way to lick this issue, maybe not by needing to use algorithmic processing anymore, but via some other ways, like maybe modeling the DNN in some other more direct and efficient way instead.

The white paper (via a junk email address)

Whitepaper - Breaking Through the 6ms Latency Barrier.pdf (1.0 MB)

@Volusiano said:

So I guess my question to the More users out there on this forum is whether you guys notice any kind of audio latency in the More that would bother you?

None. Whatsoever. Hell there’s always latency in acoustics, because the speed of sound just isn’t that fast, anyway. I’ll bet if you watched a guys lips closely enough, you could detect latency in a guy yelling at you from 50 yards away!

My $.02 from the perspective of a (retired) HPC systems engineer, after reading the tech papers and attempting to remove marketing-speak . . .

To the question of whether Oticon uses what Chatable labels as “inline” vs “hybrid”, my read of the Oticon tech papers is that its implementation appears to fit Chatable’s “hybrid” definition of using DNN for characterizing and categorizing inputs, which are used in conjunction with other technologies to select and execute traditional sound processing capabilities (albeit with greater sophistication) . . . It is “passive” in the sense that its DNN is only used to characterize sound inputs using models from which the appropriate weights/biases are derived and which close the gap between observed/presumed outputs and real/ideal outputs.

Chatable claims an entirely different approach, whereby the network is deployed “directly” to both the analysis and re-processing of sound for delivery. It is “inline” in the sense that the network is supposedly utilized throughout the analyze/characterize/modify/deliver steps. This implies a dynamic end-to-end processing model. It’s at this point that the Chatable paper becomes (IMO) unpleasantly unhelpful.
The tech paper says that Chatable is just now shopping their technology. Getting from there to a mass-produced product is a long haul that will cost a whole lot of money. For the OEM HA manufacturer, while the reward would be huge, so would the risk. I will definitely stay tuned, but there is nothing here I see that changes what those looking at HA’s now, need to think about. Come back in a couple years.

Again, just my two cents.

Edited for clean-up.