@Volusiano yeah i saw that hole, and assumed it was a mic, but you’re right, it does line up w/the internal contact.
As long as it’s not like the Phonak type, where you have to engage the contacts with the charger, which was a pita!
1 Like
That infuriating hissing sound is something I’m currently trying to figure out how to minimise on my Philips 9040, which is obviously an issue shared with comparable Oticon devices. Was really hoping that the upcoming 9050 might improve it, but thank you for looking deep into the issue and setting my expectations accordingly.
I’ve got a client with the same issue: relatively good low frequency hearing, but can always hear the noise floor.
All Starkeys AI is on thru the cloud vs being on the chip so there will be differences in the response time as well as what they do. Personally find a much clearer natural sounding signal with Real. Also find it does better in noise. This being said can’t wait to get my hands on the Intent so I can put it thru the tests
Can the floor noise be reduced by them adding settings in the software?
Two ways that I know of:
- Fine-tune the 50dB curve down a bit in the region where the audible hiss is. (Or get a more accurate hearing test done!)
- Switch off directionality (as according to the specs mentioned above by @BernhardG you could be almost 10dB better off this way, if you don’t mind hearing what’s behind you)
No, it’s almost impossible to remove in certain aids.
What is an aid known not to have this issue and how do those specs compare?
Below is an excerpt from a post I wrote on this forum when I was evaluating the Sonic Enchant 100 a long time ago. At the end of the excerpt, I mentioned how I learned what Sonic did to alleviate this issue.
2 Likes
Those who hear floor noise without directional mode enabled are just out of luck?
As mentioned by @withoutwings in an earlier post, you can use his method 1 to suppress the gain (where the floor noise can be heard) manually and this prescription override should suppress the noise floor even if you use directionality. But it comes at a cost if your hearing loss needs amplification in the suppressed range.
But usually the reason the floor noise is noticeable enough for you is because your hearing is good enough in that low range to pick it up in the first place, so hopefully the gain reduction will not have a significant effect on your audibility in that range because your good audibility will be able to pick up sounds in that range without needing much amplification there anyway.
However you should have a different program for streaming if you do the above, because streaming will need that low amplification for the low range sound content, so you don’t want that low gain amplification artificially suppressed for streaming.
1 Like
That’s a good point about the need for a second general program that works with streaming. I did a little testing with this process in the past and when the gain is near zero on the lowest frequencies, the curve is sometimes hard to fit in the neighboring frequencies. Do I ignore the fitting curve shape and just try to meet the fitting need at each frequency? Is it trial and error to identify the frequency of the hiss and then how to tweak things to reduce it as much as possible?
Below is a link to an article that talks about directional hearing aids. In the section in there titled “Low Frequency Roll-Off”, it gives an example (also shown below) showing the attenuation of the roll-off in 2 cases of directional setup where the mics are 12 mm and 6 mm apart compared to a single omnidirectional mic.
As you can see, the roll-off is fairly linear. It doesn’t really have a knee point. I think it really is up to the HA mfg to decide where and how to insert the compensation gain to combat this roll-off, so there’s really not a clear rule of thumb to follow in order to “undo” their roll-off compensation if you don’t know how they compensate in the first place. It also depends on the low frequency audibility of your audiogram → do you have almost perfect hearing in the low range, or do you have “some” hearing loss in this range?
I would say that if you have a ski slope loss with a fairly flat audibility line in the low ends and at some point it takes a nose dive, I would probably want to use that nose-dive point as the starting point to reduce your gain curve. And based on the chart above, I would not do a flat reduction from your nose-dive point backward, but maybe do some kind of a linear reduction with more and more pronounced reduction as you go back further to the lower frequency end. This probably would also alleviate your issue of having to force an abrupt gain reduction at a knee point that is difficult to fit in with the neighboring frequencies like you mentioned.
However, this is still pure guesswork, and it’s best if you keep making very small incremental adjustments at a time until you find something that works for you. If you do have “some” hearing loss in the low range, then the adjustment you make needs to take this into consideration as well, so it can get to become a bit more complicated due to the “extra” consideration.
As for the floor noise, I don’t think it exists in any single one discrete frequency that gives you a sharp “hiss” in one spot. It’s more likely a spread over the low frequency range because it’s the compensated amplification that raises the floor noise across the low range (more like lots of little hisses all over the low range). If it’s not a discrete resonant frequency of some sort due to something, then there’s no single frequency of “the hiss” that can be “zero’ed” in to snuff “it” out.
Huh? I think this is implying somehow processing is done realtime off remotely in cloud but there’s no way that can be true. Even if the model is cloud driven there must be locally available computation.
3 Likes
I concur with this. Below is a link to a whitepaper on the Starkey Neuro Processor which is the DNN based AI they use on the Genesis AI. It is a very sparse only-2-page whitepaper that looks more like a marketing brochure than a technical whitepaper with barely any sufficient details. There’s no mention of any cloud-based AI in it. Below is an excerpt from this whitepaper that clearly said that the DNN is based on physical hardware.
I would love to see some link on any information that says that the Starkey Genesis AI is cloud based.
So, does anyone have the release demos on yet?
I managed to almost get rid of the hissss, but there are drawbacks.
Perhaps first i should mention, that i use VAC+.
On the one hand, as Oticon is inventor of this formula, their permanent adjusting noise control works best with it. At the other hand, their strategy is to enable understanding of speech without unnecessarily distorting the auditory impression that the wearer is used to without the aids (as long as i hear about the same number of birds in my garden as other people, i do not care wheter they would sound more brilliant if i had more compensation). So VAC+ does sound a lot softer than NAL, the speech understanding ist not compromised (or there is enough room to enhance it further) and it’s much more comfortable to start the aids in the morning (and to forget to put them out in the evening 
).
So with VAC+ i have the smallest possible gain and more room to add anhancements.
If i compute VAC+ and i’m happy with what I get, the noise level is acceptable. But as soon as i enhance the low sounds with the basic controls for sound perception (“more detail”), the noise gets untolerable.
First, i determined, where the noise (mostly) is.
This is tedious, but by using a type of binary seek, i found that (corrected by my audiogram) the noise is almost completely in the range 2-3KHz. Note, that it does not make much sense to try single bands (there is a lot of information outside the curves!)
Then i tried to decrease the lo-compression (meaning, that the 45db-insert is close to the 65db-line) in this range.
On the other hand i found, that it almost makes no difference to enhance the range above 4KHz.
So i got something like
(i first enhanced the lo and mid-level at the complete range by 2db 'cause this improves my speech understanding dramitically ).
Now the noise was almost gone!
But there is no free lunch:
When you look at the speech-banana, you will find that 2-3KHz is nearly unnecessary to understand speech. But it is essential to recognizing music. This is something different to having a dedicated program for music listening.
If you lack 2-3KHz and you are in a quiet area with low background music, you will understand people talking and perceive everything to get oriented in space - footsteps, doors, glasses etc. But the music will be cut down to a chaotic sound that is quite unpleasant; eventually the DNR does an additional task to “assist” you. Although it’s very low, you cant’t ignore it, but you do not know what is played there. And it is really surprising, when you get near the speaker and hear now - hey, this unstructured noise is Michael Jackson, that i heard a hundred times before.
So this program is really valuable in some situations, but it is not suited for general use.
I constructed an additional program without the nonlinearity, but a low dedicated perceivable noise by adding a huge 4-10KHz boost to an unmodified VAC+basic:
At the moment this is my favourate.
I’m stopping this post now; it will get too long. But there are some additional thoughts about audiograms and directionality. As soon as i find the time, i will post them.
3 Likes
Great info @BernhardG! I would love to hear further thoughts that you have, and also share my own too when I get there; as once I’ve addressed all other issues (audiogram, rationale, tip/dome type) then this will be the final problem that I will try to solve.
Perhaps one of us (whomever posts next about it) can start a new topic on the dreaded “Demant hissss”, tag the other, and share/collect as much info as we can there on our journey to solving it?
Also for @flashb1024 and @Liviu :
Was in the audi office to pick up repaired Mores today, and got to see the Intent in person (didn’t try them on though). As for size compared to the Mores, they are slightly shorter, but also thicker due to the bigger battery, so volume-wise it probably isn’t too different. One of the videos claims “significantly shorter” than previous models; I think to use ‘significantly’ is a tad of a stretch. In hindsight, I wish I took a picture so I could post the side-by-side. And as confirmed, it is indeed a contact charger, but there is a magnetic element that helps snap the aid in place as soon as you place it in so it charges properly. And the magnet is strong enough that you can hold the charger upside-down and the aid won’t fall out. The design of the charger is bit more compact as well (and a bit taller), I personally think it looks nice.
But here’s the kicker that I did not like hearing. Unlike the previous couple models for Oticon, should you need the battery replaced, it can no longer be quickly swapped out by the audiologist – they all have to be sent in with the Intent, so I’m told at least.
5 Likes
Thank for sharing about this, @jcw11 ! After you mentioned this, I revisited the pic below and indeed there doesn’t seem to be a removable battery door as part of the casing at all. This might make some people want to hold out for a disposable size 13 battery version even more.
I guess if one has a good pair of backup hearing aids, it can lessen the impact of this issue. If I were in the market for one though, I would definitely hold out to see if a disposable battery version will become available, and if yes, how many days of use one can get out of it.
Their claim of a 5 year life expectancy can help lessen the negative aspect of having to send the aids in for battery replacement as well, but we just don’t know how reliable this new setup is.
4 Likes