Unmasking Filters

[Hexspa]

Managing masking is part of mixing. Strangely, I find it poorly covered in the tutorial realm. Specifically, I'm referring to EQ - mirrored or otherwise - and not panning since I don't really think that's valid.

As it turns out, masking is indeed well-documented. There's even a wikipedia article which shows research on the topic: https://en.wikipedia.org/wiki/Auditory_masking In it, there's one image I found particularly interesting. https://en.wikipedia.org/wiki/File:Mask ... s_sp11.jpg

The way I understand it, higher frequencies mask lower ones. What's more, lower frequencies mask more bandwidth than higher ones. This explains why you have to keep your sub tight but your high hats can be many and simultaneous.

When I add all this together, it makes me think: why don't we have unmasking filters? I mean asymmetrical bell filters which vary their Q and shape according to what part of the audio bandwidth they're in.

For example, an unmasking cut down at 250Hz will have a very low Q on the treble side and a sharper Q on the bass side. As you move the filter up, the lower-Q half of the filter can gradually become more like the higher-Q side.

In contrast, you could create mirrored filter shapes for positive bell filters. These could have lower-Q on the bass side and higher-Q on the treble so that the boosts mask less. You could even invert the whole thing to create bell filters which mask even more than ordinary.

I just thought of this last night. Maybe I should try it out first. Sometimes I don't get around to things so I'll just leave this here for now.

[ElVincente]

That sounds like a really great idea! Hope to see Vojteck input about this one.

[ElVincente]

I think I understood the opposite of you.
To me that wikipedia articles says that lower frequencies hide higher frequencies pretty fast. So an unmask filter, should have a sharp Q on the high side, and a shallow Q on the low side no ? The opposite of the picture you linked to imo.

[Tyrs]

Cool. Would be the perfect feature to be implemented in MixRevolution

[marzelli]

I guess it just isn't that much of a problem. We were able to make great mixes without it

[steve2KVR]

maybe mSpectralDynamics

[Hexspa]

I think I understood the opposite of you.
To me that wikipedia articles says that lower frequencies hide higher frequencies pretty fast. So an unmask filter, should have a sharp Q on the high side, and a shallow Q on the low side no ? The opposite of the picture you linked to imo.

The way I read the graph, and I definitely can be wrong, is that the single bar is the signal and the arches represent the volume of nearby frequencies. The top arch means the masking frequency (masker) is the same level as the signal.

So, at 250Hz, a masker at the same volume at the same frequency totally masks the 250Hz signal. As you go up, a signal at the same volume masks 250Hz less. If you go lower in frequency, a masker can be the same volume but its masking effect drops off more sharply.

The lower arches represent maskers at lower amplitudes. At 250Hz, maskers at 100% and 80% of the signal's amplitude fall off in their masking effect at the same rate when above in frequency. A masker at 60% the signal's amplitude has its masking effect drop off at a faster rate than those louder ones. Then, even quieter signals have a different slope.

I see what you're saying, though. If you're right then 250Hz masks higher frequencies more than it does lower ones. I'm not sure because Equal Loudness contours imply that we're more sensitive to the upper midrange. Since this is related, I imagine that higher frequencies will take precedence. In any case, if you're right, then you're also right about the shape of the filters.

The weird thing about this is that it all happens in our ear. From what I understand, our ears divide the audio bandwidth into zones. For signals to be independent, their bandwidth must not be in the zone of another. I didn't catch whether the article says what those zones' limits are, though.

Without doubt, great records are made without this. However, many records which we love have all sorts of technical faults. I think we can agree that technological progress, as it pertains to music, and our love for each are somewhat uncorrelated.

Come on, man. I'm trying to polish turds here.

[Hexspa]

maybe mSpectralDynamics

I've yet to crack that can o' worms, as well as many others.

[ElVincente]

The way I read the graph, and I definitely can be wrong, is that the single bar is the signal and the arches represent the volume of nearby frequencies. The top arch means the masking frequency (masker) is the same level as the signal.

Do you mean the single bar is the masking or the masked signal ?
To me, it's the masking signal frequency (a sine wave). The arcs being it's masking effect at distinct levels.

[ElVincente]

Maybe we should transpose down the sidechain of mspectraldynamics when trying to duck...
That could explain why I always prefer some dynamicEq tuned by ear to do that kind of task.

[Held]

That's an interesting idea. A built-in pitch shifter for the side-chain.

[mystran]

The way I read the graph, and I definitely can be wrong, is that the single bar is the signal and the arches represent the volume of nearby frequencies. The top arch means the masking frequency (masker) is the same level as the signal.

Do you mean the single bar is the masking or the masked signal ?
To me, it's the masking signal frequency (a sine wave). The arcs being it's masking effect at distinct levels.

The single bar in those images is the "masker" and the curves are the threshold levels that another signal needs to reach in order not to be masked. Those plots are very approximate, but the general trend is that a loud signal will mask other weaker frequencies at nearby frequencies, with the "threshold of masking" decaying (significantly) slower towards the high frequencies, such that a lower frequency is (much) more likely to mask a higher frequency than the other way around.

edit: The thing about low-frequency mud is a bit different and not directly about masking. Rather at low-frequencies the wavelengths are long so it necessarily (by information theory) takes a long observation time to diferentiate between two nearby frequencies, so it's harder for the ear to make sense of what is going on there if the signal isn't relatively simple and clean. There's also a cap (on the order of 50-100ms) on how long the ear will integrate "frequencies" before it interprets a sum of two nearby frequencies as "beating" instead.

[ElVincente]

The way I read the graph, and I definitely can be wrong, is that the single bar is the signal and the arches represent the volume of nearby frequencies. The top arch means the masking frequency (masker) is the same level as the signal.

Do you mean the single bar is the masking or the masked signal ?
To me, it's the masking signal frequency (a sine wave). The arcs being it's masking effect at distinct levels.

The single bar in those images is the "masker" and the curves are the threshold levels that another signal needs to reach in order not to be masked. Those plots are very approximate, but the general trend is that a loud signal will mask other weaker frequencies at nearby frequencies, with the "threshold of masking" decaying (significantly) slower towards the high frequencies, such that a lower frequency is (much) more likely to mask a higher frequency than the other way around.

edit: The thing about low-frequency mud is a bit different and not directly about masking. Rather at low-frequencies the wavelengths are long so it necessarily (by information theory) takes a long observation time to diferentiate between two nearby frequencies, so it's harder for the ear to make sense of what is going on there if the signal isn't relatively simple and clean. There's also a cap (on the order of 50-100ms) on how long the ear will integrate "frequencies" before it interprets a sum of two nearby frequencies as "beating" instead.

Thanks for the explanations!

We should even try frequency shifting.

Can't do it right now, but I can picture a test project, with pink noise vs low volume vocals.
To see what mix of treatments mspectraldynamic's sidechain work the best...

[Lairyboy]

An interesting discussion, and an understanding of the phenomena may lead to better mixes in some circumstances. Reducing the eq of the masking sound at a particular frequency may not lead to the masked sound being perceived as clearly (or at all) due to frequency proximity or harmonics. Instruments tend to, or can be made to sit in their own sonic spaces, though. If minor parts are not heard even after 'traditional eq unmasking' they are not adding anything to the mix and should be removed or the original masking sound is too loud at the same time as and should be reduced for the duration of the masked sound. Instances like this could be considered a composition problem rather than a mixing problem.

[Hexspa]

The way I read the graph, and I definitely can be wrong, is that the single bar is the signal and the arches represent the volume of nearby frequencies. The top arch means the masking frequency (masker) is the same level as the signal.

Do you mean the single bar is the masking or the masked signal ?
To me, it's the masking signal frequency (a sine wave). The arcs being it's masking effect at distinct levels.

The single bar in those images is the "masker" and the curves are the threshold levels that another signal needs to reach in order not to be masked. Those plots are very approximate, but the general trend is that a loud signal will mask other weaker frequencies at nearby frequencies, with the "threshold of masking" decaying (significantly) slower towards the high frequencies, such that a lower frequency is (much) more likely to mask a higher frequency than the other way around.

edit: The thing about low-frequency mud is a bit different and not directly about masking. Rather at low-frequencies the wavelengths are long so it necessarily (by information theory) takes a long observation time to diferentiate between two nearby frequencies, so it's harder for the ear to make sense of what is going on there if the signal isn't relatively simple and clean. There's also a cap (on the order of 50-100ms) on how long the ear will integrate "frequencies" before it interprets a sum of two nearby frequencies as "beating" instead.

I'm convinced that you're right.