Helix Needs a Capture Block: Hear Me Out.

[quadgear]

Sounds a bit like NIH (not invented here) and not wanting to kill small fry, friendly competition.

It won't be a feature until it is.

 

[NotApplicable]

NAM is based on WaveNet neural networks, and these are quite resource intensive, particularly for DSP hardware. This is also the technology used by ToneX, as far as i know.

TONEX is LSTM with a FIR filter in front.

 

[molul]

If Line 6 creates a capture method, just know that it's going to be put to this level of scrutiny if not higher. It would make sense that the Line 6 folks do something like this level of comparison in house before shipping it.

As one of the comments say, even though you can see differences, you could make really amazing sounding guitars with any of the three.

No need to focus on being the most accurate. Just be accurate enough and leave the rest on the player to write great music.

 

[ian_dissonance]

As one of the comments say, even though you can see differences, you could make really amazing sounding guitars with any of the three.

No need to focus on being the most accurate. Just be accurate enough and leave the rest on the player to write great music.

Unfortunately all the guys who don’t make music will spend more time trying to get views being loud on the internet about the accuracy, so you gotta be at least as good as the last thing.

 

[molul]

Unfortunately all the guys who don’t make music will spend more time trying to get views being loud on the internet about the accuracy, so you gotta be at least as good as the last thing.

Yeah, you're right. Still, I want to believe that most people don't care that much and appreciate other things.

On the other hand, I'm impressed that, at least in that video, the Tonex seems the most accurate (I thought the Kemper would always be the best). But it's a plugin, so líne6 wouldn't need to compete with it xD

 

[Mike Oliphant]

TONEX is LSTM with a FIR filter in front.

Care to elaborate? Tonex gets very good results for the hardware they are using.

 

[Lysander]

Care to elaborate? Tonex gets very good results for the hardware they are using.

LSTM is another type of neural network. I haven't played much with those, but IIRC they're in general less resource-hungry than WaveNet - and less accurate as well.

 

[Mike Oliphant]

LSTM is another type of neural network. I haven't played much with those, but IIRC they're in general less resource-hungry than WaveNet - and less accurate as well.

I know what LSTM networks are. I'm curious about the specifics of the Tonex implementation. Particularly if/how they are combining it with an impulse response in a single capture (as opposed to capturing a cab IR independently).

It has seemed like they are doing something like that, since they can kind of "turn off" the cab part of a full rig capture.

 

[Orvillain]

Unfortunately all the guys who don’t make music will spend more time trying to get views being loud on the internet about the accuracy, so you gotta be at least as good as the last thing.

Oi! I do make music!

 

[Elric]

I know what LSTM networks are. I'm curious about the specifics of the Tonex implementation. Particularly if/how they are combining it with an impulse response in a single capture (as opposed to capturing a cab IR independently).

It has seemed like they are doing something like that, since they can kind of "turn off" the cab part of a full rig capture.

I doubt it. They probably use some kind of simple algo for simulating its removal.

 

[Mike Oliphant]

I doubt it. They probably use some kind of simple algo for simulating its removal.

If they have a way of separating the linear part of a capture from the nonlinear part, that could be turned into an IR (for the linear) and a NN (for the nonlinear). It wouldn't be a true amp/cab separation - more of a separation of EQ from distortion/compression.

I've thought for a while that this could be a good way of doing captures. If you can handle the linear parts with an IR, there is less work for the NN to do.

The main problem is that standard IR capture techniques assume a fully linear system.

 

[Elric]

If they have a way of separating the linear part of a capture from the nonlinear part, that could be turned into an IR (for the linear) and a NN (for the nonlinear). It wouldn't be a true amp/cab separation - more of a separation of EQ from distortion/compression.

I've thought for a while that this could be a good way of doing captures. If you can handle the linear parts with an IR, there is less work for the NN to do.

The main problem is that standard IR capture techniques assume a fully linear system.

No, you lose information. It’s not like one is entirely linear and the other is entirely non-linear.

 

[Mike Oliphant]

It’s not like one is entirely linear and the other is entirely non-linear.

If by "one and the other" you mean amp/cab, I agree. Cabs are *mostly* linear. Amps are a combination. If you were able to separate out the linear part of an amp/cab full rig capture, it would be a combination of amp/cab behavior. If you are trying to get perfectly separable amp/cab behavior from a single capture, that's a problem. If you are just trying to divide and conquer the problem to achieve a more accurate profile at a lower CPU cost, it isn't a problem.

 

[EOengineer]

As I was narrowly avoiding a heart attack mulching my yard today, I had a realization.

L6 adds a NAM capture block to Helix…and HXFX - totally displaces Tonex and everyone else in the small profiling space.

This could be so amazing. I don’t want to throw around the usual buzzwords (disruptive, game changer, etc), but holy F would my HXFX gain major utility.

Would be a hell of a parting shot from the HX line.

If by "one and the other" you mean amp/cab, I agree. Cabs are *mostly* linear. Amps are a combination. If you were able to separate out the linear part of an amp/cab full rig capture, it would be a combination of amp/cab behavior. If you are trying to get perfectly separable amp/cab behavior from a single capture, that's a problem. If you are just trying to divide and conquer the problem to achieve a more accurate profile at a lower CPU cost, it isn't a problem.

Guitar speakers are non linear. How could a guitar cab be linear?

 

[Lysander]

Guitar speakers are non linear. How could a guitar cab be linear?

Guitar speakers are non-linear in frequency response. In the context discussed above, linearity means time invariance - i.e. they don't distort, and can be (very accurately) modeled by impulse responses.

In general, speakers only distort when pushed outside their operating limits. Which is bad news, guitar or not

 

[EOengineer]

Guitar speakers are non-linear in frequency response. In the context discussed above, linearity means time invariance - i.e. they don't distort, and can be (very accurately) modeled by impulse responses.

In general, speakers only distort when pushed outside their operating limits. Which is bad news, guitar or not

Thanks for clarifying - I think I’m on the right page now.