Dynamics matching?

[vectorwarrior]

I'd like to find a way to match the overall dynamics of two tracks. I'm not talking about moment to moment matching, MPhatik can do that. Instead I'm talking about overall level distribution matching.

As a comparison, you can match the frequency response of a sound by using the auto eq's. You analyse the target and source tracks and the plug-in can create an eq curve. Now, imagine you can do the same with dynamics.

Modern compressor has a sort of histogram view of dynamics that could be used for this purpose. If you could analyze your two tracks and then match the dynamics to create a dynamics curve in something like MDynamics, that would be super interesting. I don't think this has been done before.

Just checking, there's nothing that can do this already is there?

[Tagirijus]

First things, which came into my mind, reading your post: normalizing and / or follower modulator. I guess that you did not talk about normalization, right? .. so what about a follower modulator, which gets its input from the one track and controlls a gain for another track, or so?

Overall: I somehow guess that I did not understand your idea in the end, after all. Haha. ... sorry then. Looking forward to what other people might say as well.

[vectorwarrior]

Yeah, I guess I need to have another go at explaining it. What you are referring to is the moment-to-moment dynamics matching. What I am talking about is a long-term dynamics matching. You can use MPhatik to do moment to moment dynamics matching.... the problem is that this only works when you have identical content. What if you want to match the dynamics of totally different content?

What I am suggesting is an analysis of the overall dynamics... imagine you have a drum loop that have the perfect dynamics that you'd like to 'copy' or match on to your drum loop, which has a totally different pattern. You can't use MPhatik because the drum hits are at totally different times. Instead, I want to analyse the level distribution of A and B, then determine the difference and create an MDynamics graph to adjust A to match B.

It could utilise MModernCompressor's Level Distribution graph:
https://www.meldaproduction.com/webtemp ... r00.v1.jpg

Or as a different way to explain it, let's translate what we're talking about from 'level' to 'frequency response'... what I am talking about is matching the overall frequency response, which you'd use MAutoEqualizer's 'match' functionality with:
https://www.meldaproduction.com/webtemp ... r00.v1.jpg

What you are talking about is more like MMorph or MVocoder, where the moment-to-moment frequency differences are worked on.

Hope that makes more sense.

[Tagirijus]

Thanks for further explanation. So it would be about: "This audios level distribution is high on higher levels, some in the mid and also many on the lower volume levels ... I want this for the other material as well, which has less on lower and higher but many on mid volumes" ?

[vectorwarrior]

That's exactly it. I'm looking for a way to mimic the overall dynamics profile of sounds. I guess you could say it's like compression matching.

If we could do this, imagine what a multiband version could do to match sounds.

[DarkStar]

I am not sure how this would work. So let's try a thought experiment:
-- say we have 4 volume bands (v1 is quiet, v4 is loud)
-- and my good dynamics are analyzed to be: 20% v1, 30% v2, 40% v3 and 10% v4
-- and my bad dynamics are analyzed to be: 40% v1, 30% v2, 25% v3 and 5% v4

Would the dynamics range matching (is that a better term than compression matching?) be:
-- for audio in the v1 band, increase the level of 50% of it (dropping 40 to 20) into v2 (making that 50%)
-- for audio in the v2 band, increase the level of 40% of it (dropping 50 to 30) into v3 (making that 45%)
-- for audio in the v3 band, increase the level of 11% of it (dropping 45 to 40) into v4 (making that 10%)

[Held]

Do you mean something like this (starting at 5:40)?

You should be able to do the same thing in MSpectralDynamics too.

[vectorwarrior]

I am not sure how this would work. So let's try a thought experiment:
-- say we have 4 volume bands (v1 is quiet, v4 is loud)
-- and my good dynamics are analyzed to be: 20% v1, 30% v2, 40% v3 and 10% v4
-- and my bad dynamics are analyzed to be: 40% v1, 30% v2, 25% v3 and 5% v4

Would the dynamics range matching (is that a better term than compression matching?) be:
-- for audio in the v1 band, increase the level of 50% of it (dropping 40 to 20) into v2 (making that 50%)
-- for audio in the v2 band, increase the level of 40% of it (dropping 50 to 30) into v3 (making that 45%)
-- for audio in the v3 band, increase the level of 11% of it (dropping 45 to 40) into v4 (making that 10%)

Yeah, I don't think it would work great using the concept of 'volume bands' as you've described here as it's not granular or variable enough. I think it would need to be a continuous 'volume reassignment curve', so -6dB is boosted to become -3dB, -12dB is cut to become -18dB... or whatever causes the dynamics of the two references to match up. But maybe I'm wrong and your band approach would work better. I'm curious to know what Vojtech thinks of this. Regardless, I think it could be an interesting use of the level distribution curve in MModernCompressor, which currently is a sort of weirdly underused but interesting display.

[vectorwarrior]

Do you mean something like this (starting at 5:40)?

You should be able to do the same thing in MSpectralDynamics too.

No, this plug (like you point out) is very similar to SpectralDynamics. It doesn't do matching, just allows analysis to result in an average of the frequency response of one file to create a threshold for dynamics processing on another. It doesn't do overall dynamics matching of one file to another. SpectralDyanmics too, while awesome, is absolutely not capable of this.

Apologies if I'm not doing a very good job of explaining this, it's hard to describe!

[steve2KVR]

...0h, just to give it another shot...it would be great to have the MMC distribution curve view available in a lot of other plugins. I have a version of this set up in Mxxx...(sigh).../s~

[DarkStar]

Yeah, I don't think it would work great using the concept of 'volume bands' as you've described here as it's not granular or variable enough.

I only used 4 bands as an example. I had in mind many more, perhaps 0.1dB increments, so 600 or so.

--------------
PS I cannot get the Requested Dynamic Range to have any effect. How should I use that?

[DarkStar]

Q: How do you interpret that Level Distribution Graph ("LDG")?

To me it shows the Unquantified) % of samples that are at each volume level (depending of the precision of the volume measurement). How do you use the "Dynamics Level" red line?

For example, one test I did shows:
-- at a Dyn Level of 83%, the Dyn Range was 12.5dB,
-- at a Dyn Level of 75%, the Dyn Range was 15.0dB,
-- at a Dyn Level of 50%, the Dyn Range was 27.0dB.
What is that telling me?

[DarkStar]

PS I cannot get the Requested Dynamic Range to have any effect. How should I use that?

I've got this now. This may help others:

Let's say that the Dynamic Range is +22 dB (from -29dB to -7dB).
If I set the Requested dynamic range to +11dB I am asking for the range of the audio above that dynamics level to be halved. On clicking Auto compression the Threshold is set to -29dB and the Ratio to 2.00:1.

If I set the Requested dynamic range to +16.5dB I am asking for the range of the audio above that dynamics level to be reduced by about 25%. On clicking Auto compression the Threshold is set to -29dB and the Ratio to 1.31:1.

-----------------
I used 2 MMC's in series to see the result:

[vectorwarrior]

Just did a little test... so it's definitely not in samples as changing the attack/release time impacts the analysis... so I suspect it takes a reading of the current 'level' (the bouncing line in graphs that interacts with the threshold/ratio). I presume the analysis is a percentage of the current analysis 'session' (the time you've been analysing)... so the further the bar is the right, the higher the percentage of the material is in that dynamic range.

It's a bit of a difficult to comprehend graph, but I'm sure it makes sense to Vojtech, and again if we can compare two level graphs and create a transfer from one to the other, you could match the dynamics of two sounds... maybe?

[Faiky]

vectorwarrior, I know exactly what you mean. Matching the dynamic range (especially in different frequency ranges) is also important. You can do A/B matching of the bands by ear or eye with an oscilloscope. But tools like "Expose" or "audio metric ab" can read the dynamics as well as the mmoderncompressor. An automatching would be ingenious there of course!