I wonder how an actual good usecase for the display feature “Refinement” would be.
Without Refinement:
With Refinement:
“Hm, ok”, but what is the actual benefit of this? I don’t get it… Looks like a pretty Cheetah. But how to use it for practical purposes?
it corresponds to what I hear
realize, I’m editing a vintner in a field working vines
what is a waned and an unwanted sound?
hitting/ rubbing the mic I want to reduce/remove;
crew chatter, I want it gone
wind noise…well reduced
RF buzz and sporatic wireless drop out…it’s gotta go
the rest? is wanted noise
refinement greatly enhances the clarity for me…and these transients are critical to my current work
yes, I don’t understand the animal skin so well…and sometimes the “unwanteds” are just lurking in the jaguar print soup…what I can visually correlate to audio in the spectral view is helped with visual clarity
automatic tools (modules) are no replacement for the human editor at this time…similar to “slapping a limiter on it”…it just sounds better mixed by a human editor…at this time
I use refinement on max all the time, to be able to tune in the FFT size as sharp as possible. And to find and select small detailed areas.
I don’t really understand why one would want to lower the settings, and use the dull spectrogram.
And besides, the sharp spectrogram looks nice and trippy to the eyes.
I find that for most of my uses both refinement and resolution work better if not too high.
When they are above 70% refinement and 3x resolution I start seeing too much of the stuff that don’t matter, and lose focus from the real problems.
When they are a bit lower I find it easier to differentiate stuff and spot the most common problems.
this is me too 100%!
…and 98% on the time view is set to straight down
but I am a bonafide newbie to SL
I’d really enjoy to watch you work =-D
hehehaha! well, maybe I don’t know what I’m losing.
But for real, what do you actually do inside all those leopard spots? I feel like I’ll just end up selecting whatever all around them anyways… Never needed THAT much precision. Maybe I’m a grunt with no finesse?
Edit: just went and done a quick test. Took a small selection of a noise layer that had stolen a bit of dialogue I wanted. Grabbed the harmonics selection tool and carved the bit out using both 3x/70% and 4x/100%, copying them to a new layer. They had small differences in harmonics, but nothing drastic.
When summing them back to the dialogue, I found that the 4x/100% selection, while apparently more refined at first, took too much noise in the shape of harmonics, and left me with a result where I would want to EQ the high mids back in a bit. While the less refined selection took less “shaped noise” in, and left me with a more mild result, where I didn’t feel the need to EQ the high mids (which apparently were below the noise floor anyways).
So I guess there will be cases where more resolution and refinement will help, but definitely not all the time, so the 3x/60-70% usual setup remains the standard for the time being.
right on! Thanks for the details Henrique!
I will try what you describe. I have stopped using harmonic selection tool becuz artefacts (as I shared in another post)
I do find some unwanteds lurking in the animal hide at times…especially when working with music…but for film NR too
That’s very interesting! I must try that!
What was the height of the Harmonic selection since catching all that noise? 2 or 3 pixels? And how many harmonics? 5, 10, 35, 50?
If so, it would be interesting to know if the 1 pixel height would catch as much noise.
And I believe the number of the harmonics should be set to fit the material, so not to select way up in the treble register if not necessary.
What artifacts have you been finding with the harmonic selection tool?
I’ve found that if it stop the selection on the wrong place and cut, I get a click. But I haven’t been able to put my finger on what exactly makes that happen, I just go a little longer on the selection when that happens.
This conversation got me thinking about why and when exactly did I settle for those 3x/70%? I mean, I definitely don’t change those settings frequently, and hadn’t tested different ones in ages, nor did I exhaust the possibilities before settling. Then I remembered!
(if I’m not mistaken) I read somewhere - an old manual version? a post here? - by @Robin_Lobel himself, that higher resolutions are not directly correlated to better results or precision, and unless you really need it for something specific you are better off sticking to the default (is it 2x?)
So I think then I settled for a middle of the road. A bit higher than default, but not too grainy.
Reading the SL11 manual, it does say something in that direction, but I couldn’t find exactly what I believe I have read.
Maybe Robin could shine some light on the subject? I’m very curious.
BTW, I have just learned (or re-learned) that refinement is purely a visualization setting with no bearing on the result of any processes. Refinement, in the other hand, does influence results.
I’m intrigued by these controls too. I will investigate more when I have the time. Definitely got a itch about this hehe!
If I remember correctly, it was around 20 harmonics, which is the ballpark of what I found to correlate to the highest harmonic I could visualise on the spectrogram regarding this specific bit of dialogue. The problem is that this height was not really visible throughout the part, ducking below NF at various points. And it was in these points that the lower resolution didn’t grab as much as the higher resolution.
Regarding the pixel height, I really can’t remember! It could be that 1 pixel height would yield a better result in the high mids, but couldn’t it also mean that it would leave too much behind in the lows? I honestly don’t know.
The differences were subtle, so I wouldn’t say “ALL that noise”, it was a bit of more pronounced, slightly harsh hi-mids that were poking out of the general feel of the noise reduced dialogue, and the lower resolution was giving me something more similar to the rest of the dialogue (which was unmixed with Unmix Noisy Speech and whose noise layer didn’t show any significant remains of the wanted signal).
That was really the only relevant difference I found between them. The lower part of the spectrum sounded and looked pretty much the same.
Since this, I have been experimenting with changing the resolution for some tasks, but lets see what the budgets allow in terms of tinkering with what is already acceptable and within budget heheheh!
Please both of you do tell me about anything you find in this respect. This program is awesome!
thread is here
https://forums.steinberg.net/t/harmonic-selection-artefacts-when-paste-to-another-layer/928278?u=ctreitzell
Do you mean resolution affects results of selection?
From SL11 Operation Manual:
"Resolution: The Resolution parameter sets the resolution of the FFT analysis. More subpixels are calculated, but it does not necessarily means more accuracy. The Refinement parameter sharpen the spectral display so you can see more clearly the exact frequencies and timings.
NOTE Increasing resolution and refinement requires more computing power."
I guess, that in order to fully understand the nature of those artifacts, one would have to read a good mathbook about the fine intricacies of FFT and eventually another good book for the black magic of numerical mathematics and its concrete algorithmic implementation.
I actually start to wonder about the dependencies of (graphical-)selection, FFT-Size, Refinement, Resolution and the chosen FFT-Window. Not forget about the magnification in x- and y-direction.
See here:
Which datapoints in the spectrum are chosen? I don’t know.
(Of course, I can work with it. But some questions remain open. And yes, it’s clear, SL is not meant for scientific purposes. But heavy science goes into SL, though 
)
I can’t speak for @Robin_Lobel nor Spectralayers but I used to do architectural work(blueprints) in matlab and from my experience I believe the resolution and refinements act as an overlapping extension of FFT.
For example, if you understand synthesizers and dsp and how that works, there is a basic problem called aliasing(especially in the higher frequencies). There are techniques to combat aliasing by oversampling. Oversampling(depending on the sample sizes) also known as anti-aliasing can be achieved by multi-sampling(whether x4 or x8 or x16), however oversampling uses more computational power because it requires quadruple(x8 or x16 or x32 and-so-on-and-so-on) the number of samples.
Synthesis: for the most part, yes, I grasp using (subtractive) synthesis…depending on the type wavetable/samplers being another different animals.
DSP; not beyond effect routing…I wouldn’t know anything about coding beyond GM and SysEx control system (which I understand quite well)
I’m a musician…electric guitar being my natural instrument
I’ve been dabbling in audio for a long time too
I have this FFT reading on the to do list…there are A LOT of things ahead of that for me, tho…like wood burning stove and firewood, cuz woodburning season is just around the corner. Suffice to say I know very little about FFT smp size…so far, all I can relate this stuff to is kind of a quilted patchwork to represent audio in the digital realm…I see it as kind of an image sprite with variable positional controls…even if I do read up, I got no idea how much math would sink in and how much that knowledge might aid me. It certainly isn’t going to get all my editing I must finish to get done!
Marc, I guess you’re speaking of the artefacts I linked to? I don’t really think about where they are coming from…I just know it happens. I was previously making harmonic selections and then cut>paste multiple times from the same selection to another layer…the multiple paste operations increased the artefacts…as I explained in that thread. Part of the reason I posted was to make other newbies aware.
Nowadays, I might make a harmonics selection and paste said selection to a dummy layer for a visual guide and then use the brush selection tool to select>cut>paste because I get no artefacts with the brush tool. Is it more work? Well, no…because cleaning out the harmonics selection artefacts can take a very, very long time and using brush tool, I’ll just be done.
Sometimes I’ll purposefully use harmonic selection and clean the artefacts. It’s not so easy to make harmonic selections beyond the area you want to select because once the harmonic selection tool falls off the Master Rank, you can’t “back up” and must re-select (afaiaa). Then, going back through using erase to remove/ reduce the artefacts. I don’t like using this method because it brings unknowns into my editing. There has been no official comment about the harmonic selection artifacts from . If it is a known issue, Robin hasn’t said anything about it that I have seen.
I will say, I’ve never even tried the Magic Wand tool in SL…maybe once. What is it selecting? I suppose you can listen back to the selection. I’ll give it a try…what is it based on? Amplitude?
Operation Manual says:
“Select an area by automatically detecting shape contours where clicked.”
what are shape contours?
Google sez:
" 1. : an outline especially of a curving or irregular figure : shape. the sleek contours of the car. The map shows the contour of the coastline…"
oh my, clear as mud
I need to just crack on…SL11 has been the best tool for my film audio that I have ever tried…I’ll just keep using it
I’ll just update that I’ve been working at 4/100 today with direct on-top view of spectrograph, getting my recent editing that is not the same as my current workflow and staying mindful of this discussion.
I’m mainly moving what I term “Speech Rems” (remnants) to another layer (labelled Speech Rems). This is speech left in the noise layer after running Unmix Noisy Speech. I was using Harmonic selection tool, but I found issues as I explained before. So I use mostly Selection Brush (round at between 20-30px depending on selection (getting smaller as freq rises). I seldom select over 1K-2K unless a consonant has a hole. To select higher freqs, I use I larger selection area…50-200px.
I bring this up because I find it accurate enough for my purposes. And it does work well.
My workflow is essentially
-unmix noisy speech
-create a layer for unwanted nz
-create layer for Speech Rems
_then get to work moving whatever hasn’t been automatically unmixed to the destination layer I expected. Unmix Noisy Speech works very well IMO/E
A lot of inhales and sibilance find their way to the Unmix Noisy Speech noise layer, and I don’t worry too much about moving those.
I would find it pretty tricky to work the same way with a blurry spectrograph.
I’ll read the rest of the thread later, but just wanted to give you the heads up that you can in fact back-up on selections, just press ALT + click and drag your cursor!
You can also add harmonics to a selection by holding shift and clicking on a higher harmonic with others already selected.
It selects the partials.
