Sometimes there’s jobs where dirty source material needs to be cleaned up fairly heavily and it can be hard to ride the line of too much/too little.
We have the Noise Spray tool, but I’d definitely like something a little more automated.
Noise Regeneration:
- Ability to set threshold for where noise is adding (only filling holes)
- Dynamic response with attack/release to blend well with remaining noise, and or respond to source dynamics, and or a dynamic “mix level”
Harmonic Regeneration:
- Levels for Tone/Synthetic vs Noise, ie the ability to choose/blend between actual tones or noise levels that form harmonic points
- Attack/Release for both Tone and Noise based on source material dynamics
- Level target (so only harmonics are generated from intended source)
Ability to send these processes to separate layers (additive layers, instead of subtractive layers) for mixing. The user should be able to customize which layers are created:
- Noise Regeneration Layer
- Harmonic Regeneration Layer: Noise
- Harmonic Regeneration Layer: Tone
I’d also appreciate Noise Reduction module to have a variable harmonic preservation amount (ability to also go to a separate layer)
Please also see:
Eraser Limit Improvement - SpectraLayers - Steinberg Forums
Improved Hum Reduction and Other Noise Reduction (Compete with RX) - SpectraLayers - Steinberg Forums
edit
Perhaps, this is also the key for better AI results. From what I can tell, SL has mainly focused on 1-stage processes - ie, focused on the Removal/Reduction/Separation Stage, but perhaps realistically, there needs to be a 2nd Regeneration stage after the first process… and even if it is slightly “synthetic”, the end-result might still be better than without.
Believe-it-or-not, the unmix levels feature works perfect for this, the problem with it is that processing is offline (so you cannot use it to audition audio in real-time).
I hope the next update of spectralayers directly does unmix levels in real-time.
I use Unmix levels a lot, but it’s not really tackling what I’m describing in this thread, at the very least, it’s not as efficient as what I am imagining/describing.
I’m considering the term ‘Regeneration’ as creating something that technically isn’t there to begin with, because something had to be removed that was in conflict.
Unmix Levels, those conflicts are removed to a different layer - but that still leaves holes in the keeper source material, and the conflicts still exist in the separated.
Yes, you can go in and edit the conflict layers and then reintroduce them, but then you have holes in the conflict layer.
@Robin_Lobel
For example when using Unmix Noisy Speech - you get two layers - the speech and noise layer. But in the noise layer you now have holes from the unmix process.
Often I would like to use the noise layer somewhere else - but the holes create audible issues.
What I would like to see is the ability to use the speech layer (or any other layer) as a mask layer (like in Photoshop) to create a selection. And then a healing tool to fill in the gaps by cloning/interpolating/growing the noise layer surroundings of the gaps into the gaps.
Hope that makes sense.
1 Like
@awesomeaudio
Can you upload here an example of what you’re talking about?
Or private message me!
Robert, indeed, I asked for this long ago; it just seems so logical. BUT, consider systems which are lower spec than your production powerhouse. If SLP tries to create a selection based upon a layer’s audio, load on RAM will create a spinning wheel/ “not responding” message for lower spec machines…and potentially crashing. I might be wrong, yet that was the re-framing of my thinking 18 months ago.
So, what about a time selection which then renders a selection based on amplitude of audio within said time selection? Then, users could select areas that are smaller than the entire layer and might mitigate the memory issues.
But, yeah, I’ve wanted this from day 3 of using SLP.
I manually fill the holes with clone stamp in my workflow. It is very time-consuming, so only do it when it is really needed.
To get pedantic, I’d say the selection would be more of a “key” than a “mask”.
@Robert_Niessner working on it ! 
3 Likes