[anigbrowl]

What’s the solution? Multi-Band Compression (MBC), a technique that’s been used by the $6 billion hearing aid industry to solve this specific problem.

An MBC uses intelligent design instead of a one-size-fits-all method. With the right data about your hearing pattern, it can mash the full sound into your range so that you get all the information you need.

Audio engineer here. That is patently untrue. MBC is a super-useful technique and is indeed helpful for mitigating hearing loss in relatively transparent fashion, but it does not and cannot bring sounds from outside someone's audible hearing range back within it. It will dynamically rebalance incoming audio in inverse proportion to the degree of hearing loss within a set of frequency ranges, but many kinds of sensorineural hearing loss involve the death of cilia cells (the tiny hairs thatvibrate at particular frequencies, much like the bins of of an FFT) which can result in a total loss of perception at or above certain frequencies.

http://en.wikipedia.org/wiki/Sensorineural_hearing_loss

To 'mash the full sound into your range' requires a technique known as frequency shifting, but that's problematic because it destroys the harmonic relationships of the incoming material and sounds disorienting, at best.

In any case, I think the illustration of the bear on the tricycle is absurdly simplistic and makes me wonder to what the degree the pp designers really grasp the underlying concept. A much more appropriate parallel would have been to show an image with a severe Gaussian blur, which more closely parallels the actual experience of hearing loss in terms of both empirical measurement (higher frequencies tend to be more severely attenuated in cases of induced hearing loss) and subjective experience (blurring hinders edge detection, which is analogous to transient detection in audio, and which has a large role in speech intelligibility.

http://en.wikipedia.org/wiki/Gaussian_blur

If you're struggling with hearing loss, then you should really, really consult an audiologist, work out the basis of your hearing loss (which is sometimes as simple as impacted earwax), and work out a treatment strategy. If you're suffering from degenerative hearing loss then listening to overly-compressed music could actually accelerate it, and listening on headphones or earbuds (many of which bias the sound for increased impact) could also contribute to the problem. It's a truism in the pro audio world that most people are awful at self-measurement and tend to over-equalize in the absence of proper experimental control protocols.

I apologize for the rather negative tone of the post; I appreciate the people at SoundFocus are trying to provide people with something useful and helpful at minimal cost, by leveraging the pretty good audio hardware in their phone. However, hearing loss tends to be a one-way thing, and I think that offering a product to that market without a clinician on the team is a bad idea. There's a lot more to being an 'audio ninja' than understanding the fundamentals of DSP.

[function_seven]

Yeah, the article confuses things with the wrong definition of dynamic range, then goes on to explain MBC using the term "range" correctly.

MBC can "mash the full sound into your range" if "range" means dB range at each freq. band. But since the author previously (ill)defined dynamic range as a frequency-related term, the reader reads that passage and thinks he's referring to frequency shifting instead.

[pistle]

I was about to question everything until I read this. Thanks AE. My understanding of MBC is more like a crossover (or mix of low, bandpass, and highpass filters) network followed by compression per band within each section of the frequency range.

Now, I want to go test out what it would be like to 'compress' frequencies. Something like a notch filter that shifts nearby frequencies around the target frequency away into regions above and below. It adds noise, essentially, within the compressed range, but maybe it's tolerable and is useful for someone with a narrow band hearing loss. It could potentially be interesting musically.

Maybe such a filter exists, but I am not familiar with it.

[anigbrowl]

If you're interested in frequency shifting, Harald Bode was the leading engineer in this area. You can read a gentle introduction here: and if you look around there are some VST plugins that emulate the Bode designs.

I haven't tried using this for precision stuff - over a small range it might well improve intelligibility at the expense of only minor distortion. I tend to reach for it when I want to give sounds an extra weird dimension, it sounds somewhat orthogonal to the normal harmonic distributions we're familiar with.

[wglb]

Looks like you were going to paste a link but it didn't stic. . . .

[w4]

>I was about to question everything until I read this. Thanks AE. My understanding of MBC is more like a crossover (or mix of low, bandpass, and highpass filters) network followed by compression per band within each section of the frequency range.

This is exactly right. Basically, you separate the audio into arbitrary frequency bands, and then apply compression to each band to control its volume independent of what is going on in the rest of the spectrum.

I was incredibly frustrated by reading the article, since their explanation of multiband compression was incredibly misleading. I get what they're doing and why multiband is helpful (it sounds like they're basically bringing up the volume in the parts of the spectrum where the user's hearing is less sensitive than healthy hearing would be), but that was a poor explanation of how multiband compression works.

[pistle]

Well, that's probably why. Somewhat obvious. If you put something through with content in and around the range, you get a notch filter with a resonant hump... not all that interesting.

[stolio]

I thought the bear picture was maybe the best part of the article. Compression is tricky, multiband even more so, and the type of multiband compression they're using is even harder to wrap your head around (there are two ways (that I know of) to do what they're doing but I doubt they want to talk about which one they use.)

It's a great layman's explanation, but if you have a better one I'd love to see it.

[bentcorner]

I think the point of the bear picture isn't that you can't see half the bear, but rather that (to continue the visual analogy) you can't see half of the bear's colors. How do you fix something that you can't perceive? You can change the missing colors to something that you can see, but you end up distorting the original image.

In the soundcloud samples, if I can't hear anything above a certain frequency, making them louder isn't going to help. You can drop the frequency of those things, but my guess is that it's going to sound pretty ugly. It would be interesting to listen to a sample that has everything above a certain frequency pitch-shifted downwards.

[stolio]

Here's how I took it: if you're just losing sensitivity at a particular frequency then you may only hear sounds in the 40-100dB range, below it's too quiet to register and above it's painful. That's a lot of information to lose but you can smash the 1-100dB range into the 40-100dB range. If you choose to you could even smash the 1-60dB range into the 40-60dB range (or pick whatever numbers) and leave everything above that relatively untouched. This is a fairly common sound engineering technique to fill out a sound without destroying its dynamics.

So if you picture a scale next to the bear picture from 1-100, then the bottom part of the bear is what's beneath the (effective) noise floor for that frequency. To extend the analogy to multiband compression you'd have maybe 10 bears next to each other, each missing different amounts and each needing a slightly different smashing to lift the bottom of the picture into the visible range.

edit: I think people are assuming that the frequency content of the bear picture corresponds to the frequency content of sound (they're all signals, right?) but to me it's a much more basic analogy. To do it that way you'd have to be turning up the soft reds or something to that effect, but rods and cones being what they are we don't lose vision in a comparable way to how we lose hearing so I don't think there's a good, intuitive visual analog in that sense.

[varunsrin]

hey anigbrowl, OP here.

You raise some great points - it is indeed impossible to use an MBC to bring sounds back into a user's range of hearing if they have lost all sensibility at that particular frequency. However, the loss of hearing at a particular frequency is not binary - it tends to start with a reduction in dynamic range at that frequency, as the cilia start to get worn out / destroyed.

So if you have loss @ 3 KHz, you don't often completely lose all hearing, but your dynamic range which normally is 0dB -> 100 dB (over-simplification here) might now be 30 dB -> 100 dB.

What an MBC will do here is compress the range at that frequency band, so your 100dB of range is now 70dB of range.

[anigbrowl]

I get what you're aiming at, and I applaud what you're trying to do - I just think that you've over-simplified a complex topic, to the point of creating quite a bit of confusion, going by this thread. To quote Einstein, “Make things as simple as possible, but not simpler.”

[diydsp]

There are numerous interstitiated mechanisms in hearing. I don't see why MBC in particular would be valuable, other than amplifying quiet parts automatically, which could in fact have the opposite effect.

However, I think there is potential for many other mechanisms to be developed, such as automatic filtering, to eliminate masking in frequency and time domains.

The question I have is why other companies, such as Apple and Spotify, don't simply add this DSP technology to their software. What can SoundFocus do that can't be copied? Proprietary algorithms?