Recreating electric guitar DI signal?

Nowadays when recording electric guitar I’ll often record its DI signal from the guitar in addition to the amped sound.

I have some old guitar tracks that I’d really like to have DI versions of, and figure SpectraLayers is the tool for the job. I’m at best an occasional user of SL and was wondering how those with more expertise would approach this? The audio is pretty straightforward just amp overdrive and probably(?) a distortion pedal. There aren’t any time based effects like delays or chorus cluttering things up.

Any advice is appreciated.

So you’re trying to remove distortion added by the overdrive and distortion pedal, and get back to the original guitar signal ? Sounds like Process > Clip Repair would be a good start.

Thanks. That makes sense as a starting point but unfortunately didn’t help. I ran it at a variety of settings on a test file with no real audible impact. Then I looked at the file in Wavelab and realized the waveform was not even close to clipping anywhere.

Here’s what the audio (a 3 note lick) looks like in both SpectraLayers and Wavelab. You can see the overtones clearly. I’m wondering if that is the straight signal and the stuff in between is the distortion? Not really sure how to interpret what the spectrogram is showing.

To be honest, I would be very surprised if there was any current DSP that could remove willfully added distortion that typically messes with the entire frequency range of the original signal.

In the future, this may be possible via Machine Learning - there’s an interesting research paper from earlier this year:

But even that approach almost sounds more akin to drum replacement, where it’s more about creating entirely new audio, by analyzing an original signal together with some other information about the desired output.

Or in other words: One can analyze and recover the originally played notes including string/pitch bends and then create effectively a new guitar track as it might have sounded when played with a specific pickup combination.

Which also adds the interesting issue: A hot humbucker can already bring some distortion to an otherwise pretty clean pre-amp. So even recording that via a DI gives more distortion than a squeaky clean low output single coil.

With Machine Learning one day we may even be able to “change the pickups” after the fact.

1 Like

@raino There’s a form of clipping, even if you don’t clearly see it in the waveform. In SpectraLayers for the process to be effective, you have to set a clipping threshold below what you see in the waveform. I would try -18db for instance, base on your Wavelab screenshot (btw, there’s a waveform visualizer in SpectraLayers too).

indeed - and mix ML with audio resynthesis to fill in the gaps and almost anything seems possible.

1 Like

producers/remixers: :metal:
historians and judges: :scream:
philosophers: :yin_yang: :person_shrugging:

1 Like

I’m not sure it you saw or listened to this link:

but they checked out Cedar at the NZ police department. Yes, we are living in a post truth world.

Our ears are not the best judge of what is true.

1 Like

FYI Cedar’s main client is law enforcement and has been since the start (at least that’s what they told me during a showcase of their hardware). We are merely playing around with the toys that Law Enforcement has abandoned :slight_smile:

1 Like

Finally got a chance to play with this some more. Setting the threshold to -18dB actually increased the distortion level.

Then I had a duh moment and realized I had Audio Files of DI Guitar which would let me see the end result I’m after. This pretty much shows a bunch of clear harmonics and nothing else. So I tried using the Harmonics Selection Tool and then copying the selection to a new Layer. This worked reasonably well. Initially I selected 10 harmonics which sounded fairly dull (like the tone knob on the guitar turned down). Next I tried 18 harmonics and while that brightened the sound it started bringing in some distortion. So now it’s just fine tuning the number of harmonics.

One thing I wanted to try but don’t understand how to is to narrow the frequency range of the fundamental selection making it ‘thinner’

The musical problem I’m trying to solve is that the different guitars are playing interlocking lines that fit with each other, but they were all recorded with the same rig & guitar sound so they are difficult to hear independently. But with a DI-ish signal I can give a guitar line a whole new & different sound.

Here’s the DI signal from a recent Project.

And here’s the 18 Harmonic version of the distorted 3 note lick shown above

1 Like