IK Multimedia has introduced AI Machine Modeling – a new artificial intelligence powered system that they say is their most revolutionary amp modeling technology yet.
AI Machine Modeling software lets you, without any custom hardware, model the sound of any amp, cabinet or combo, plus pedals like distortion, overdrive, fuzz, EQ or boost. They say it can do this accurately enough that the results are “virtually indistinguishable from the real thing”.
The system works by analyzing your dry signal against the processed signal, and then creating an algorithm that models your hardware rig. Complete rigs are entire signal chains including amp + cab, and different types of FX pedals such as overdrive, distortion, boost, EQ and wah.
It’s unclear at this point if this technology will have useful applications beyond guitars. If it could model effects chains applied to synths, it could be very useful for synthesists.
IK has a variety of audio comparisons at their site. They say that AI Machine Modeling is ‘coming soon’.
“It’s unclear at this point if this technology will have useful applications beyond guitars. If it could model effects chains applied to synths, it could be very useful for synthesists.”
A couple of applications come to mind:
1. you could add the dirt layer of a vinyl recording to make a sample sound like sampled from a record (no, these vinyl emulations don’t cut it)
2. you could “sample” artifacts from vintage samplers
3. you could imprint the characteristics from, say, an isolated Michael Jackson acapella, modeling the original signal chain onto your own vocal recrodings
4. wild experiments in sound design: record a vocal underwater, apply characteristics to other signals
are we still claiming neural nets are AI instead of just pattern matching? ooo, let’s bring back expert systems too; nothing like a listening to the product of if/then/else statements to set your mind at ease.
Looks like you noticed a pattern.
Neural nets are most definitely AI. In fact they’re the core of modern Deep Learning. Check out convolutional and recurrent neural networks, they’re considerably more robust than static decision trees.
For more general info IBM has a good primer here https://www.ibm.com/cloud/blog/ai-vs-machine-learning-vs-deep-learning-vs-neural-networks
No, neural nets are objectively not a product of deterministic if/then/else statements.
If it’s not too much to ask, try to stay positive when posting. This kind of comment really spoiled my coffee this morning.
I really do think all this stuff is great. But… given that the new tech is here, why does 21st century music suck so very badly?
New video, making the 1953 Fender Bassman with AI Machine Modeling https://youtu.be/R6kbE6NgIYc