As I can see, Oticon More and Whisper are not absolutely ideal - there is something newer Chatable Launches AI V3.0 Edge On-chip Inline DNN | The Hearing Review – a MEDQOR brand
The Chatable article doesn’t give much details except for boasting that it’s the best thing since the invention of the sliced bread. I don’t recall if anyone had ever said that Oticon More or the Whisper are absolutely ideal anywhere, though.
I tried to download the Chatable whitepaper but it requires me to give away my email address to someone named Giles. I tried using a fake email but apparently that was a no-go. The last thing I want is more spams in my Inbox.
Do you have a copy of the Chatable whitepaper that you can share somewhere without requiring an email?
At least Oticon doesn’t require me to give away my email address to get spammed on in order to read its whitepaper. Whisper has no whitepaper as far as I can find anywhere.
4 Likes
Most start ups disappear, very few ever really go to production.
1 Like
Maybe it was a prank? But Giles Tongue is real man https://www.linkedin.com/in/gilestongue?originalSubdomain=uk#search
Thanks, it looks like Giles Tongue is a co-founder of Chatable according to LinkedIn. Nevertheless, I’m in no hurry to give him my email address just to get their whitepaper. I’ll just wait to see if they become legitimately successful first before investing my time reading their whitepaper anyway. Maybe by then they will grow out of their insecurity and won’t require anyone’s email address before they release their whitepaper.
3 Likes
The spiel doesn’t say “latency in doing what”? What functions are being directly executed in real time, and what AI is being deployed?
At least Demant has been transparent about it: I know what sounds Oticon is processing the in real time with its AI. (Would you believe seagulls, anyone?)
I live 700m from the bank of a saltwater estuary. Seagulls are an important, if secondary, part of my life, so I’m a direct beneficiary of Oticon’s cutting-edge DNN seagull technology.
Any day now, when my ON app is actually working, they’ll be notifying me that they’ve finally released a firmware update that targets the damned crows!
I’m-a-waitin’!
Creating a junk email address for such uses can be handy.
2 Likes
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.
1 Like
@Volusiano:,When are you going to break down and buy some More1s? They’ll rebalance your 🦤 seagulls 🦤 big time!
2 Likes
Haha, Jim, I’ll break down and get the More when one of my OPN 1s breaks down first. 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.
3 Likes
I hear you … when the deal is done, we’ll induct you into the Secret Society of the Syber Seagull and teach you its secret motto!🥸👍🏻
1 Like
I’m quoting myself here just to offer a paraphrase of what Oticon says in their literature, based on my own personal experience.
Hope you get yours soon. Waiting in line too.
The audioologis clinics aren’t so greedy.
1 Like
You probably won’t get the 3 year warranty or low-cost loss replacement from them. That alone is probably worth the price difference considering that if you buy locally, you can probably get your near-end-of-warranty new replacement for the pair (as long as your audi is OK with processing this for you) so that you get the chance of being able to get double the longevity out of the HAs; not to mention the full programming service the whole time from the audi.
I’ve bought a used pair of OPN 1 for my brother in late 2018 from eBay for around $1100 with the ConnectClip included, before the OPN S was released in 3/2019. They came in almost brand new condition with all the original packaging. The shipping labels of the packaging suggested that the seller got it from the VA office.
2 Likes
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.
5 Likes
The white paper (via a junk email address) 
Whitepaper - Breaking Through the 6ms Latency Barrier.pdf (1.0 MB)
4 Likes
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.
4 Likes