Improvements in the behaviour of frequency/harmonics selection

SpectraLayers
1

Hi all! Hi @Robin_Lobel !

I have used a lot of frequency/ harmonics selection tool lately, and I have found a few things that would improve its usability vastly, in my opinion.

Since the result of the selection changes quite a lot depending on the settings, it would be great if we could alter the selection settings AFTER selecting, so that it results the same as if we selected with prior configuration.

Very frequently I find myself selecting what I think is a fundamental, only to find out that in fact the fundamental is hidden, and the lowest note is actually a first harmonic (common in shouts etc.), and vice-versa; or maybe that I have chosen too few harmonics, or too many, or that I should try with more pixels… you get the Idea.

So it would be great if we could just select any frequency, and then go about changing the settings until we find the optimal setup, and the selection follows suit. And here I include changing between “frequency” and “harmonic” selection modes.

Another thing that would be SUPER useful, even if the above never comes to life, is the ability to chose how many frequencies we want below the “master”. Many times I find that the harmonic series of a given sound is structured so that the lowest note present is a 2nd or 3rd harmonic, but if I select it as such with the master control, then I also select a bunch of noise under it. So it would be nice if we could have that option too.

I believe these modifications would greatly improve the speed and ease of use of this tool and avoid a lot of “selection>oops no>change>select>oops no>change> select> ok, finally.”

And also, since this type of selection is always controlled by the X axis, it would be great if the selection would be always locked to the cursor position while we hold the click down. As it is, if you move further than where you wanted, you have to erase a part of the selection, and that leaves some mask behind.

And yet another one, is that many times I can see very clearly where a frequency is going, and want to follow it, but the selection disagrees with me, straying up or down, It would be great if we had a way to “force it” to catch on to another stream and follow that. Maybe a modifier key that would bias the detector towards the cursor position? or two modifier keys that would bias it Up or Down. I think this would also save me from a lot of time fighting the selection path and ultimately changing to another tool to finish the job.

If you think these are good additions please give them your support, and if you have any objections please also say so.

And to Robin, thanks for your time, as always.

All the best!

1 Like
2

The selection system within Spectralayers is already good as is. I do agree that the editing function of the selection system could be improved starting with more functionality in axis points. I already suggested for the selections to be able to be like the region marker/spectral marker function where you can select any where (around the selection) and it creates an axis point with the ability to lock and unlock that axis point.

Yeah! That’s been a problem for a while but I’ve gotten used to it and worked around it. I understand why it does that though, and that is because the horizontal frequency intersects aggressively with the vertical frequency. In this case, this could be achieved by redefining what a partial is and what Spectralayers considers a partial.

1 Like
3

Add me to the want list here
+1

as far as the artefacts from paste, tho…just select well beyond and delete with Time Selection can work here

4

Vertical frequency?!? I do not grasp what you speak of
Time is the vertical component;
Frequency is the horizontal component

are you talking about an amplitude difference?

What I understand by using spectral layers is when making a freq selection with freq/harm selection tool;
the tool follows the highest amplitude vertically depending upon the user’s selection criteria (ie 1px…2px…10px, etc)

Therefore, the selection tool just selects the next highest amplitude of that (horizontal) freq lane…and if it is a 70db dropoff; user doesn’t get a warning for that

I gotta say, I have stopped using the harmonic selection tool because of the leading trailing artefact of a pasted copy of said selection. As I already said, I will use harm select for longer passages and/or as a guide to use other selection tools.

5

@ctreitzell

There’s a transient selection tool (vertical axis which are usually percussive sounds) and frequency/harmonics selection tool (horizontal axis which are usually tones and harmonics sounds). The 2 of these tools seems to be both operable to vertical and horizontal axis and nothing in-between. There’s the magic wand tool that is close to selecting whole partials but not close.

What I’m suggesting is either the tools functions have to be redefined or a whole separate tool would have to be created to select whole partials. The reason why I believe the frequency selection tool behaves this way is because there is a sharp transient(vertical) along the horizontal axis and therefore considers that a transient.

1 Like
6

Well written…
Agree with everything in this post…
+2

2 Likes
7

JK,
Please don’t take my interest in clarifying details as an argumentative…I’m simply trying to understand other users’ experiences and share my own :slight_smile:
I don’t follow your logic. You are saying the transient tool selects horizontally and the freq/harm selection tool selects vertically only? cuz then you say:

What considers “that” a transient? The trailing edge of the freq/harm selection? The leading edge? Both edges of selection?

IME, the trailing edge of a freq/harm selection (after a paste action) creates a downward “transient”- I find excess lo freq energy on a harmonic selection which appears to travel vertically down (Transient, if you like) from the lowest partial and is not very obvious that it is happening; as compared to the leading edge of a selection which creates a vertical “sound-energy” which shows vertically across “all” freqs upwards, above the trailing edge selection and leaves a transient sounding click and said vertical “anomaly” is visible as well.

The trailing, lo-freq “transient” doesn’t make a “click” sound because it is lo freq energy…and likely below the majority of playback systems and typical human audible recognition.

“along” the horizontal axis? I don’t understand this…I think you mean vertically across the horizontal axis. I would think “along” would mean a horizontal “event” (narrow band event)…whereas across might go the opposite direction (vertical) and sound as a transient event across many frequencies…I’m just trying to clarify…

I’m not going to assume I understand these event “anomalies” because I don’t…so I can’t speak to analyzing what is happening with the selection tools…I can only report my experience.

btw, I was thinking that amplitude in regard to spectrograph is z-axis :slight_smile:

I will need some time with Magic Wand to see what I find that doing which I’ll get to over the next few days

8

@ctreitzell

Yeah! You’re right! @Robin_Lobel should step in and explain further.

1 Like
9

@henrique_staino some interesting suggestions there - I have some improvements planned for SL12 regarding this tool, I’m adding your ideas to the list.

@ctreitzell @Joey_Kapish the frequency selection tool consider at each step the next “pixels” to the right (or to the left if you’re going left), and goes with the one with the highest amplitude. Which is indeed not always the best solution, as a crossing transient, or some noise burst can make it go off track. The frequency tracking algorithm could certainly be made more robust, along with the other suggestions here.

3 Likes
10

Thanks for the clarification, it is deeply appreciated :smiley:

Obviously, I’m a semantics fanatic…I’m just interested in clear communications and deeper understanding of functionalities for myself and any others seeking clarifications.

“Selection going “Off-track” from a transient or noise burst”;
Freq/Harm selection seems to go off-track with a big drop in amplitude as well and then, yeah, appears to look for next highest amplitude, so if trying separate human voices occupying the same space, becomes a very manual endeavour. Nevertheless, it is possible to do a pretty great job of separation with the tools SL11 already has.

Something which might help SL be more capable of more effective separation might be some kind of user modelling where timbres are captured by selection and then applied to other source material…Unmix Multiple Voices does attempt to do that IME

None of this “explains” the “transient” generated on the right hand side from a paste operation of a harmonic selection other than @Robin_Lobel saying the freq/harm selection tool :

thanks, Robin, for your open and honest communications with us in the forum; you lead by example and, again, I personally deeply appreciate your unparalleled support of your fantastic software; thank you so much for sharing it =-D

2 Likes
11

Thank you so much Robin
Thanks your dedication, I am sure SL will be the best software in the world

1 Like
12

Yeah! About that!

Believe-it-or-not, the find similar feature works on multiple voices but it only works mostly on vowel sounds.

A feature like that is definitely doable on a host of a ton of GPU’S servers. The key is efficiency.

2 Likes
13

I haven’t tried FIND SIMILAR IIRC

I did try Unmix Multiple Voices which was better done manually in my case…I have to try find similar too…

14

Unimix Multiple Voices is 100% broken for me. It never ever separates different voices correctly, no matter how dissimilar they are. Not even close, never. I have tried every way possible, it just never gets anything even remotely close to correct or useful. And I mean completely off. Like it just gives me different blocks of chopped sound with whataver are in them with absolutely no correlation or apparent regard to how any of the voices behave.

2 Likes
15

@henrique_staino are you properly registering individual voice signatures first ?
If so, could you share an audio sample with quite dissimilar voices where it fail to identify them properly ?

1 Like
16

Hey @Robin_Lobel !

Yes I am. this is an old battle of mine, heheh! I believe I have talked about this in this forum before.

Absolutely, I will collect a few examples and send you both the original files and the resulting separation.

You’ll see that even when the voices are alternating in time it struggles to keep them each on their own layer.

Please bear with me cause I’ll take some time to be able to prepare the examples ok?

Thanks a lot as always!

Enviado via Proton Mail para Android

-------- Mensagem original --------

1 Like
17

Robin, I have a recent job where my main contributor has an interruption of passers by and they have a conversation because that main contributor has planned to give them a tour of local waterways. The result I got with SL11 is excellent, IMO but it did take over two weeks work in SL11 for me to get there.

I think we’d be well served by someone like @Phil_Pendlebury doing a really in-depth video of Unmix Multiple Voices following your instruction.

I did register the various speakers with Unmix Multiple Voices, yet I could have easily done something wrong along the way. I can share this job with you if you like…via email?

My layer naming convention might be a bit goofy for you…BUT I typically have the original file in my jobs as a safety

1 Like
18

Dear Robin
am saying that since the beginning.
The separation voices have issues.
I am not able to separate correctly.
Could you please explain more clear what you mean for PROPERLY REGISTERING INDIVIDUAL VOICE SIGNATURE FIRST???
Seems to me that you have better procedure than mine; fantastic; please give me the entire process to obtain the correct separation voices
Thanks

1 Like
19

@Gennaro I checked your previous messages on the forum but couldn’t find reports from you about the Unmix Multiple Voices module ?

1 Like