The No.1 Website for Pro Audio
 Search This Thread  Search This Forum  Search Reviews  Search Gear Database  Search Gear for sale  Search Gearslutz Go Advanced
MAnalyzer: tuning for best accuracy Channel Strip Plugins
Old 15th July 2018
  #1
Lives for gear
 
Ain't Nobody's Avatar
 

MAnalyzer: tuning for best accuracy

EDIT: Wanted to add that I've since come across this video which addresses some of the questions, though others remain:


While I had assumed that tuning for maximum accuracy would be a rather straightforward cpu tradeoff, this doesn't appear to be the case, and I'm scratching my head about a few features in particular.




SUPER RESOLUTION MODE
The manual says:
Quote:
Super Resolution Mode activates a special processing algorithm, which provides high resolution even in the low frequency spectrum. Using standard FFT algorithms you can increase the FFT size to get better bass resolution, but this also slows down the response. Super-resolution mode keeps the quick response in high frequencies as they are naturally quicker, but also highly enhances the bass spectrum resolution. It requires additional CPU power
Sounds straight forward, and the video shows it as being a quick fix that just provides more accurate low frequency results, but in using it, I've been shocked at how different it is... sometimes causing 10dB swings in the low end, and there's the rather obvious issue of the tilt:

Pink noise with Super Resolution mode DIS-engaged


Pink noise with Super Resolution Mode engaged


Green is peak, orange is average, and both graphs on a 3dB per octave slope, so the pink noise should be essentially a straight horizontal line. It is with the SRM disengaged, but WTF is going on with it engaged? There's a very obvious tilt going on there... and unless the pink noise generator in Logic is broken, it's just not correct.

It's actually MORE perplexing on actual material. Here's a snapshot of a song. Again, green is peak, and orange is average.

Actual song, SRM DIS-engaged


SAME song (believe it or not) with SRM engaged


Look at those peaks in green. Yes, it's the same sample. What's going on here? Is there some other setting that must match, are the SRM results garbage, or am I missing something?




ANALYTICAL SMOOTHING
The manual says:
Quote:
Analytical smoothing switch activates a more complicated smoothing algorithm, which provides more accurate results, however it may require much more CPU power. Unlike normal smoothing this method doesn't change the proportions of frequencies with higher magnitudes. It is useful mostly for technical analysis and for most musical signals it is often better to use the default smoothing method.
Again, it sounds at first like a straight ahead compromise between accuracy and CPU consumption. I'm fine with that, but which is it? If it provides more accurate results, then why is it often better to use the default method... or is this just referring back to the CPU issue?

The video clarifies a bit about it not diminishing harmonics like the typical smoothing does, but again, I'm not exactly clear whether the comment about it being better to use the default is only about CPU, or about something more meaningful like our perception not taking the harmonics into account the way analytical smoothing does.

Again, the goal here is accuracy from the perspective of human hearing.




TIME RESOLUTION
The manual says:
Quote:
Time resolution improves the time resolution, but lowers the spectral resolution. This is typically useful for more scientific analyses, where the signal is moving quickly and you need to follow its movements quickly. This is often advantageous for sonograms with very high FFT sizes.
OK, so maybe I just need this at zero. I would be interested, tough, to hear from anyone who has it set differently that can share why they are using this control, and how it increases accuracy for their intended task.




WINDOW TYPE

The manual doesn't even attempt to explain this one, but basically says leave it on the default (Hann) unless you have a specific purpose for changing it. That's fine. Melda is full of stuff that's just got more controls that are needed for any given task. Love that. The video touches on it briefly, though I still see no particular reason to use anything but the default.

Really just out of sheer curiosity, and since I'm already here asking other related questions: Who's using this in other modes, and why?

Last edited by Ain't Nobody; 16th July 2018 at 04:46 AM..
Old 15th July 2018
  #2
Lives for gear
 
Ain't Nobody's Avatar
 

Should've mentioned: The FFT size for graphs above was 4096. Higher settings look slightly closer for the overall shape of the low end on real music examples to SRM mode, but there's just no escaping that tilt:

Pink noise on 3dB/octave graph with 2262144 FFT size and analytical smoothing:


Pink noise on 3dB/octave graph with SRM on and analytical smoothing:


In addition to whatever it may be doing for the sake of accuracy, it certainly looks like SRM mode also causes a 1dB/octave (roughly) tilt from 6k or so down to 20hz (actually, the peaks in real world material appear to be skewed even further than that, though it's harder for me to judge which reading is more accurate in that case..)

Is there some other setting that can counteract this, or is SRM just not usable? If so, is the right answer essentially to use very high FFT size, and analytical smoothing? I'm fine with the slow refresh, CPU hit, and whatever else. I just want accuracy.

Last edited by Ain't Nobody; 15th July 2018 at 10:47 PM..
Old 16th July 2018
  #3
Lives for gear
 
Ain't Nobody's Avatar
 

Really hoping for some qualified feedback here. The whole purpose in diving into this was to get an accurate read on the high and low end, and I just did not expect the kind of swings I'm seeing.

Increasing the FFT size alone from 4k to 8 or 16k with no other changes can cause 10dB swings in the sub readings making me doubt what I'm seeing.

Here is the exact same clip with three settings. Each has analytical smoothing, and smoothness of 7% to make it a bit easier to read.

Typical 4k FFT with 4x overlap (SRM off)


Settings cranked to 262k FFT size and 512x overlapping. I'm assuming the issue here is that it's just taking so long to process that the infinite average (orange) never settles down. Could still be the case, but this was taken after nearly 5 minutes of a 6 second clip repeating on my 8 core machine. Nothing against plugs allowing absurd settings. Maybe that's just the case here.


And... same settings as second image, but with SRM engaged.



Is it or is it not the case that higher FFT size is always better, and higher overlap is always better? If not, I need some sort of rule of thumb to find an optimal compromise. If so, do I just need to step it down a bit to make sure I can wait out the full processing period to get a stable picture? I'm relatively sure at this point that analytical smoothing is always a good thing as it seems to do a better job preserving peak values at higher smoothing settings.

And is there some way to salvage super resolution mode? Note that yet again, it appears to be bending the image around 5-6k, and tilting everything down to the left. It doesn't do this in the video, yet it does 100% of the time in all of my tests.

Last edited by Ain't Nobody; 16th July 2018 at 05:34 PM..
Old 16th July 2018
  #4
Lives for gear
 
Ain't Nobody's Avatar
 

Some of this stuff should just have a right answer, yet I see none anywhere. Analytical smoothing... does or does it not better approximate our hearing? It's not just making the top end brighter as I might expect from the description of it preserving the harmonics. It's completely changing the shape of the curve, so saying it's more accurate, but then saying it's better to leave it disabled for music is just doubly bewildering.

SRM seems unusable as far as I can tell, but is there some other setting in order to avoid the signal being bent at 6k? It's so consistently bent in dozens of tests now that I just assume that it's essentially doing one kind of processing above 6k, and another below... and that the bent graph is just a fact of life. Just not using it for now unless someone can point out something I've missed. I'm tempted to just fix the bend in prefiltering, but verifying that works properly beyond test signals would be difficult.

I understand the rest comes with the territory, and is full of compromises, but I'm finding it surprisingly difficult to just find any solid reason to go with ANY particular settings as default. I wouldn't consider this too much of a problem except I've spent two days now going through hundreds of permutations, and am consistently amazed at the differences.

Unfortunately, they're often not consistent either. It would be simple if I actually believed, for instance, that raising the FFT size increased bass accuracy (as both the manual and video say), but I can see that's not the case when I raise it once, and the bass jumps, then raise it again, and the bass drops way down. There's no progression towards any kind of ideal. They cannot BOTH have been more accurate than their predecessor.

Please tell me there's some compromise that's not random, but that strikes an optimal balance between concerns for specific reasons.
Old 17th July 2018
  #5
Lives for gear
 
Ain't Nobody's Avatar
 

I was under the impression this was one of the most widely used analyzer plugs.

Are folks here using something else? If so, same core question... Ignoring (within reason) CPU concerns, what's an optimal setting for accurate measurement of tonal balance?
Old 17th July 2018
  #6
Lives for gear
 

I use the default setting to give me a quick overview on the bus. You can slow the response time to make the display a little less jittery and make the results a little more overarching. But don't depend on any analyzer to pinpoint issues. Use it as a "big mistake preventer" and rely on your ears for the rest.
You might also run pre-recorded material that you quite like the sound of through it and take note of the general shapes that you see. Pay particular attention to the way that the top end droops. I think you'll find this to be key.
I'm quite happy to have this living on my main out but I don't use it beyond casually checking unless I have some problematic peak. I generally find that on material coming into the studio now, as opposed to stuff going out. The advantages of old fart experience, I suppose....
Old 17th July 2018
  #7
Lives for gear
 
Ain't Nobody's Avatar
 

Sure, and I'm doing all that, but I see no valid reason to NOT try to make it as accurate as possible (CPU issues and other practical concerns aside).
Old 17th July 2018
  #8
Lives for gear
 

Quote:
Originally Posted by Ain't Nobody View Post
I was under the impression this was one of the most widely used analyzer plugs.

Are folks here using something else? If so, same core question... Ignoring (within reason) CPU concerns, what's an optimal setting for accurate measurement of tonal balance?
Says to update to a paid version to be able to resize. Guess that´s why Voxengo´s SPAN is more popular.

In use, they all visualize differently and it comes down to individual experience. Never have been really happy with analyzers in EQs either. Till recently with the analyzer in the FLUX EVO channel strip. Reveals quite different peaks than others. And what is more important a 1-2 dB cut can rebalance a file immediately. Think the developer's background in neurosciences helped to develop a quite different approach.
Old 17th July 2018
  #9
Lives for gear
 
Ain't Nobody's Avatar
 

Quote:
Originally Posted by lllubi View Post
Says to update to a paid version to be able to resize. Guess that´s why Voxengo´s SPAN is more popular.

In use, they all visualize differently and it comes down to individual experience. Never have been really happy with analyzers in EQs either. Till recently with the analyzer in the FLUX EVO channel strip. Reveals quite different peaks than others. And what is more important a 1-2 dB cut can rebalance a file immediately. Think the developer's background in neurosciences helped to develop a quite different approach.
Not hung up on using Melda. It just seems to be the most extensible from what I've seen, so I figure it can be optimized more than most.

Haven't used that, but love some of flux stuff. I think I remember them having an expensive analyzer.

I understand it's a trade-off between concerns, uncertainty and so on... so not looking for a miracle here, just an optimization of those concerns.

What am I optimizing for? Short clip (3-10 seconds) average energy per frequency. If there's then some way to adjust that a bit to give a bit of weighting to short loud noises (another thread), then great. As for overall accuracy, though, I'll settle for just being able to trust that if it says there's more of this freq than another, that it's the case, and it matches our experience (equal loudness issues aside.)
Old 20th July 2018
  #10
Lives for gear
 
Ain't Nobody's Avatar
 

Really hoping someone knows something more specific about the analytical smoothing. The materials are, unfortunately almost deliberately unclear as to whether it is or is not a better match for our perception. Unfortunately, it's not as straight forward as simple bit of extra tilt or something due to the preserved harmonic peaks. It can, in some cases, at least, change the curve pretty radically.

Same for increasing FFT size in increments. Bass goes up with one increase, down with the next, up with the next... so clearly it is NOT the case that simply increasing FFT size increases low frequency accuracy. If that were true, there would be a more logical progression.. Really hoping there's some insights to be had there.
Top Mentioned Products
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump
Forum Jump