Oticon New Product: Oticon Reveal

Perhaps they work for a particular current President?

Sorry, yes I was referring to Sphere. I forget that there is the regular non-sphere.

No, they’re still pushing brain hearing.

Yes, the marketing departments need some reigning in. Resound’s 88% more accurate claims are also based on studies that didn’t do spectral analysis. If you don’t do spectral analysis you aren’t really assessing the performance of devices using DNN technology which makes comparisons meaningless.

More detailed information about the various technology levels is available here.22 I still won’t be able to recommend below tech level 2 for this product when competing manufacturers offer much better performance in noise for the price given the relatively poor input dynamic range at tiers 3 and 4.

It seems there a balance point beyond which removing contextual cues impairs the brains ability to understand speech. This makes sense as the brain has its own speech enhancement circuitry for separating speech and background noise which requires specific information in order to work effectively.

That doesn’t mean that Oticon is optimising this balance point better than anyone else. This is just the story that they’re telling to help promote their product.

The only way to find out how the technology works , it to try them out . It has to be in natural environment . When you hear the marketing from the manufacturer , it is not enough .

My understanding is that the strategy is to achieve that balance point, which you say is a “marketing story”. The whole point of separation of the dual AI is to achieve this balance. At least that was my understanding from Matthew Allsops video and their general blurb.

The thing is, if you look at the graph I posted above speech understanding improves beyond the balance point. I think their strategy will be more complicated than just achieving the balance point as listening effort/fatigue also play a role and when you’re tired you’re less able to concentrate so removal of more background sounds may be beneficial. It also depends on cognition. People with poorer cognition need higher SNR’s to achieve the same level of performance. Children have less mature brains and poorer language knowledege thus they require a much higher SNR to achieve the same level of performance. We’ll see how Oticon present the information to clinicians. I don’t think there’ll be anything revolutionary with this product based on what I’m seeing so far.

@Hearing_Potential said:

More detailed information about the various technology levels is available here. I still won’t be able to recommend below tech level 2 for this product when competing manufacturers offer much better performance in noise for the price given the relatively poor input dynamic range at tiers 3 and 4.

Wow, 113 dB dynamic range in Oticon Reveal 1 and 2. I wonder what it is like in the case of Phonak.

@Hearing_Potential said:

It seems there a balance point beyond which removing contextual cues impairs the brains ability to understand speech.

The larger SNR loss, the more environmental sound should be removed.

It uses a size 13 Li-Ion battery that is integrated and not user replaceable.

So, that Johnson and Healy paper was interesting, but…I’m not sure interesting enough to truly be useful for oticon marketing. Granted, I skimmed, but my initial take away from the paper was that, on average, people with hearing loss need a higher signal to noise ratio than normal hearing individuals to be able to clearly identify BOTH speech and noise. Not revolutionary, but I think the actually interesting point here is: even in relatively QUIET situations, individuals with hearing loss need a higher SNR for both speech and environmental sounds to be clearly identified, so perhaps there should be certain amount of environmental noise reduction on at all times, which is probably what oticon is trying to apply (and imply) if they are leaning into this study (e.g the hearing aid is trying to constantly maintain this balance).

The big limitation that pops out of the study is that they had ten people with hearing loss and 9 of them were mild/moderate. So their number, ~5.4 dB SNR, only really applies to that population. Are we to imagine that oticon has done internal studies to look at this more carefully? I bet…not much, but who knows. Are we to imagine that this idea is on their radar now and they are working with it in mind? I certainly would if I were them. The research tells us (based on my skim, please allow for error) the range of SNRs required for different types of sounds to be identified in the presence of speech (while still understanding the speech) for certain hearing losses. You can well imagine that if AI is moving towards being able to apply a specific SNR per sound per hearing loss, that could be useful.

Now I would also expect that there are certain hearing losses where their criteria, 95% accuracy for recognizing both the speech and the sound, will simply not exist and they’ll have to turn to maximizing speech alone with the possible exception of emergency noises.

Are any of the current AI implementations per hearing loss? It seems not and we can’t really know as these algorithms are a black box.

Exactly, in Oticon Intent 1 there was DNN 2.0. Are there DNN 3.0 in Reveal 1?