DLC86
Roadie
- Messages
- 581
here's the training software and the white paper about the neural network used:
M-Vave Tank-G | 60€ pedal with AI-based amp capture
- Thread starter DLC86
- Start date
here's the training software and the white paper about the neural network used:
northern_fox
Roadie
- Messages
- 453
That's neat! Although doing an A/B without a cab/sim made me cringe haha. Will be interesting to see more
DLC86
Roadie
- Messages
- 581
Yeah, they could have also made a better routing for the comparison, cuz I think there's some difference between the model and the pedals due to the input impedance of the pedals (I hear a bit of tone-suck).
Did you take a look at their white-paper? Cuz they say their approach is sort of a hybrid between Wavenet and LSTM which should result in a method that has the best of both, but there are not a lot of details about it in the paper
PS: at that price I'm tempted to buy it just to test it and make a video about it
northern_fox
Roadie
- Messages
- 453
Yeah, they could have also made a better routing for the comparison, cuz I think there's some difference between the model and the pedals due to the input impedance of the pedals (I hear a bit of tone-suck).
Did you take a look at their white-paper? Cuz they say their approach is sort of a hybrid between Wavenet and LSTM which should result in a method that has the best of both, but there are not a lot of details about it in the paper
PS: at that price I'm tempted to buy it just to test it and make a video about it
Haven't read yet but I'm gonna guess based on the price that it can't possibly be strong enough DSP to match tonex/NAM. However, I'd love to be wrong!
DLC86
Roadie
- Messages
- 581
Yeah, thought the same thing
Mike Oliphant
Groupie
- Messages
- 43
I can't remember exactly what the hardware inside the tonex unit is, but it isn't super beefy.
Mike Oliphant
Groupie
- Messages
- 43
I read the paper, and there isn't much information there other than stating that it is a new custom network architecture.
It is certainly plausible that they've come up with an architecture that has better quality/CPU tradeoffs for lower powered systems than either LSTM models or small wavenet models.
It would be great if they would make their training code public. It would probably benefit them in the long run, as I don't think they'll get much traction with their own proprietary format.
It is certainly plausible that they've come up with an architecture that has better quality/CPU tradeoffs for lower powered systems than either LSTM models or small wavenet models.
It would be great if they would make their training code public. It would probably benefit them in the long run, as I don't think they'll get much traction with their own proprietary format.
Sascha Franck
Rock Star
- Messages
- 5,493
The first example doesn't even remotely sound like the pedal.
