Say you have a stereo drum overheads track.
You run SL Unmix Drums on the track.
You now have Kick, Toms, Snare, Hihat, Ride and Crash symbals separated.
You solo the Crash symbal and focus on one hit. It looks like this:
Notice the vertical dark gaps. They are Snare drum hits and they are assigned to another layer, namely the Snare layer by SL, which is totally understandable.
What is the best and proven way of getting the Crash symbal layer fully restored, filling the gaps?
In the Crash layer, can you copy larger areas with the Frequency Repair tool successfully? I can only manage to draw a super thin horizontal line at a time.
What is your preferred way?
1 Like
Iād say if you donāt have a clean cymbal hit (or any sound) to fill in the blanks I donāt see how else it could be done without bringing back what was removed
hence, regeneration is what is needed
The time it would take to manually try to fill in the gaps would be quite a lot for one hitā¦how about doing a whole song! Yikes
you might get somewhere with clone stamp if you zoom waaaay in, yet I doubt youād be happy with the resultsā¦I should think clone stamp would be better than freq repair tool, tho
The answer to this depends on the structure of the original mix. Often, overheads are stereo, but snare is mono. Similarly, hi-hat is typically mono. That said, overheads pick up everything, including snare and hi-hat, so itās still a complex problem which directly depends on the number of point sources (mics) supplying mono components of the kit.
If you use M/S, you should get good separation of the mono from stereo. You should do this before you unmix the drums, then use other techniques such as EQ and/or side-chain-filtered gating or compression to lower the volume of non-cymbal sources in the side/difference signal.
In other words, you can use tried and true, long standing techniques (before you resort to unmixing) to optimally prepare the audio for unmixing. By doing some of the work beforehand, you should significantly reduce the capture of snare into the cymbals channel when unmixing.
2 Likes
Thank you both.
@digitaldiggo Iām trying to understand your comments, but cannot make them fit my scenario.
Mid / Side is not used. Are you suggesting that I should somehow convert it and work in M/S regardless?
Itās a stereo recording of drums in the overhead position. Iām not recording it. Iām mixing it. I do have the snare drum available on its own mono track (point source). I donāt have the cymbals available, as they werenāt micāed separately.
1 Like
I know what you want - you said so. I dont know how the drums are micād, but it seems youāre saying the entire kit is micād via a pair of overheads set up for a stereo downmix (your recording)
Todd explained why bits of the cymbals are āmissingā when unmixed - the cymbal sound at those points in time is masked by the snare, maybe also by other high pitched drum sounds.
My suggestion is to encode the stereo file into M/S.
The snare, kick and other centre sources should be much lower in the side (difference) output than the mono (sum) output. This is a way to reduce the impact of the snare on the cymbals, because the cymbal sounds will be quite a bit louder than the snare sounds in the side (difference) output. Thereās more āāstereo informationā from the cymbals than the snare.
The question Iām more interested in is āwhy do you want to restore the so-called āmissing informationā from the unmixed cymbals channel?ā - I dont get why thatās your goal, unless you want the mixdown to include soloād cymbals? The information you are concerned about is masked by the rest of the drumkit, so at those points in time the cymbal hit is impossible for the unmixing algorithm to discern and it concludes the entire sound at those points is something else (such as the snare).
My suggestion allows you to exaggerate the stereo information just for the cymbals which may assist the subsequent unmixing algorithm to more accurately discern cymbals from snare.
Also, donāt assume your only unmixing choice is Spectralayers. ReStem by WaveMachineLabs is far more powerful at unmixing drums than Spectralayers. ReStem ONLY does drums (with excellent results), whereas Spectralayers is a spectral editor with free 3rd party unmixing models incorporated into the program. Spectral editing is very different to unmixing. As Todd suggested, it could be a lot more time consuming to use resynthesis/interpolation to create a facsimile of each affected cymbal strike in your recording by using spectral editingā¦.and even if you do that, the result is likely to be masked again in the mix by whichever parts of the drums are currently masking the cymbals at discrete points in time.
Thatās why I offered the M/S suggestion.
If you want to try the spectral resynthesis/interpolation option, you should use vertical interpolation, where (for each affected cymbal strike) you tell Spectralayers to source the data (used to resynthesize the mssing data) from immediately above and below the missing frequency element in the Edit>Heal tool.
Also note: I use a different spectral editor for resynthesis tasks, not Spectralayers.
1 Like
@Henrik_Hjortnaes I think @digitaldiggo is suggesting to use a workaround of unmix M/S applied to your stereo OH and possibly might return something closer to the ballpark that you might be looking for 
Iāll offer up an example: I have a colleague who is much more of a vision editor who has been working on a film for many years. He was struggling with the audio and I suggested and showed him SL. He purchased SL and go to work on NR for his film. Afterward I talked to him and he said he was getting great results with unmix crowd noise for his NR needs. There arenāt big crowds in his film, yet using unmix crowd noise was delivering the results he was looking for.
you donāt have a single, clean crash hit?
hmmm, and another thought
by layering the one crash hit on multiple layers, you could time shift the various layers to cover the holes in the rarefication and edit out (select and delete) the overlaps; that might be a lot faster to carry out and you will be able to un-pick your work if you utilize trash layers too
again, it isnāt going to sound perfect without the missing parts in their correct velocity and time domain
The cymbals cover the full stereo field, from left to right.
The remainder of the kit is point source i.e. not always centre mono, but close enough for this task.
If the OP provides a 30 sec sample, I can demonstrate.
But the real question remains: why attempt to ārestoreā a sound which is naturally masked by other elements of the kit? The cymbals are there, but just notched at certain frequencies because the notched frequencies arent audible above the rest of the kit. It seems a bit like tilting at windmillsā¦
1 Like
yeah, not for me to say 
I was just looking to help best I could
