Do we really need another Tube Screamer analysis?

[dano12]

Well, this one is different, and a fun read also.

https://teddykoker.com/2020/05/deep-learning-for-guitar-effect-emulation/

[GibsonGM]

Very interesting, good post I'm curious how the models react to dynamics that are all over the place...soft to loud.   That's one thing that has ALMOST always been missing for me in typical modeling amps/DSP units...they didn't "feel" like they reacted the same way to playing dynamics.  Maybe it was due to some difference in output impedance or something (?) vs the real effect.  Of course that complicates this exercise by an order of magnitude, ha ha...you'd have to bring an amp into the mix.

[PRR]

""To maintain reproducibility, we set all of the knobs on both the pedal and audio interface to 12 o'clock:""

In a non-linear path, that is not sure to cover all tones.

Perhaps more interesting is the link to Fraunhofer Institute's guitar etc datasets.

[Digital Larry]

That was like those dreams I have sometimes where I'm taking a final but I haven't gone to class all quarter and have no idea what any of it means.  Although I do get the idea that this is how some of these new fangled modelers like the "NeuralDSP Quad Cortex" accomplish their goals.

[Mark Hammer]

I mentioned in passing here, on other threads, that when I attended Summer NAMM in 2018, I had a lovely 10-minute chat with Christoph Kemper, of the Profiler and Access synths.  One of the things we discussed was the capacity of digital modelling to capture/replicate the behaviour of drives and fuzzes with the authenticity that it replicates just about all other types of effects with.  I expressed the view that we aren't quite there yet, while Chris felt we were.  That led to me asking him what particular clipping pedal he found most difficult to emulate.  His answer surprised me.  Perhaps he was being sarcastic (he does have very good poker-face kung fu).  Perhaps he didn't hear my question correctly (NAMM is a very noisy environment).  Perhaps I didn't make out all the details of his reply, for the same reason.  But his answer was: the Tube Screamer.  I found this odd because one of the goals of the TS is the consistency of tone it provides, or at least aims for, across the entire fretboard.  If anything should be easier to model and emulate, you'd think that would be the one.  Go figure.

[garcho]

Maybe he's having a hard time modeling the hype

[GibsonGM]

Maybe he's having a hard time modeling the hype

Hey hey, you just CAN'T model mojo, baby!   

[rockola]

Subject is misleading; there is very little if any analysis involved in throwing a bunch of signals at a deep learning black box.

"To maintain reproducibility, we set all of the knobs on both the pedal and audio interface to 12 o'clock:"

In other words, a great approach if you have found the holy grail settings and are willing to glue down all the knobs. Now excuse me while I turn all the tuner knobs on my guitar to 12 o'clock and glue them down.