With all of the mutterings with AI, and obviously cool stuff going on with guitar tones, I though of something that would perhaps fit within the Melda 'wheelhouse' and could be just as cool:
Allow the user to input two recordings of identical sounds, one 'dry' and one 'wet', and generated a new simulated reverb. Sure, there is already IR sampling, but it would be so much more convenient (and probably accurate enough for most people) to be able to just have a dry recording of anything, and then record the same thing in a space (or throught an effect chain) without worrying about impulse clicks or sweeps.
Neural Networks are *Awesome* at being to take an input and turn it into an expected output. This means that training the model would just be throwing lots of inputs, and comparing the outputs with a known IR. (The training could also compare it directly with digital/software reverbs).
If MConvolution.. (or something similar/new) shipped with this capability, a growing 'trained' model as part of the product, and then users could further 'teach' the model with their own inputs (and optionally submit then back, so long as there were checks to prevent 'poisoning' things), this could build a solid community around reverb, and add far more flexibility.
I fully expect the 'not possible' or 'not going to do it' but if I didn't mention it, I'd be forever thinking about it, as I'm sure it is possible to some degree and whilst may not produce exact (or even 'close') copies, it would provide a great creative capability... computers should help where they can.
Idea for MConvolution (or similar)
-
- KVRAF
- 9578 posts since 6 Jan, 2017 from Outer Space
Recording a sweep is a pain, but if you record just a click (some call it impulse), you get a usable impulse response immediately and most likely a better one that an AI could get out of two recordings…
-
- KVRer
- 28 posts since 18 Feb, 2021
I'm a massive fan of guitar amp capture/profiling.
That said I don't see the point of the idea of dry signal + wet signal and getting the difference between the to to make a reverb impulse.
far more useful is something that can actually separate and then subsequently remove the reverb from a signal.
Making IRs (impluses) of a reverb or space is already a well established process. I'd be supportive of something that simplifies the entire process so that more people could use them.
That said, once you have MTurboReverb - most algo reverbs are covered, and MConvolutionMB plus any IRs is a pretty powerful too and between those - there isn't a huge gap left as far as reverbs go.
A better process to create acoustic IRs would be welcome - along the same concept of Input dry + Input MIC'd IR that represents the difference. I still haven't been able to get the expected results using Melda tools yet in that area.
I'm never in a spot where I have the dry signal and the same signal wet with verb that I don't have the tool that just created the Wet version. The more interesting verbs are going to be the really authentic long ones that an IR isn't going to replicate anyway.
That said I don't see the point of the idea of dry signal + wet signal and getting the difference between the to to make a reverb impulse.
far more useful is something that can actually separate and then subsequently remove the reverb from a signal.
Making IRs (impluses) of a reverb or space is already a well established process. I'd be supportive of something that simplifies the entire process so that more people could use them.
That said, once you have MTurboReverb - most algo reverbs are covered, and MConvolutionMB plus any IRs is a pretty powerful too and between those - there isn't a huge gap left as far as reverbs go.
A better process to create acoustic IRs would be welcome - along the same concept of Input dry + Input MIC'd IR that represents the difference. I still haven't been able to get the expected results using Melda tools yet in that area.
I'm never in a spot where I have the dry signal and the same signal wet with verb that I don't have the tool that just created the Wet version. The more interesting verbs are going to be the really authentic long ones that an IR isn't going to replicate anyway.
-
- KVRist
- 123 posts since 5 Dec, 2019
Is anyone up for a guitar/bass challenge in Melda?
https://www.meldaproduction.com/guitararchive/guitar
They're looking for Dry/Wet material for machine learning.
I'm not a good guitar player so I just watch
https://www.meldaproduction.com/guitararchive/guitar
They're looking for Dry/Wet material for machine learning.
I'm not a good guitar player so I just watch