MoMa : First Machine Learning Plug-in by Analog Obsession

[AnalogObsession]

Here is first Machine Learning plug-in 'MoMa'!

It's fully passive mojo box with Glue & Mojo modes based on Ethan Winer’s design “Mojo Maestro”.

To read more about this project, please visit the project page.

DIY Mojo Maestro by Ethan Winer

Also, check Ethan’s website for wonderful informations!

ethanwiner.com

Special thanks to Ethan Winer.

FEATURES

- MODE

1. G-E: Glue - Even Harmonics

2. G-O: Glue - Odd Harmonics

3. M-E: Mojo - Even Harmonics

4. M-O: Mojo - Odd Harmonics

- AMOUNT: Will set amount of Glue or Mojo

- MIX: Will simply blend DRY and PROCESSED signal

Resizable interface. Simple "Bottom Right Corner Handle" to resize. 50% to 200%.

*Resizing Tip: When you duplicate plug-in, it will remember current state of GUI size. But if you load new instance, it will be opened at default size. To get rid of this issue, simply follow these steps;

1. Open plug-in for the first time

2. Resize GUI according to your screen resolution or you liking before setting any knobs

3. Save this state as default preset with your DAW

Now, you will be able to load new instances with same size.

(If you want to go default size, simply double click to resize handle.)

Version 1.3 - 19.3.2024

- Reduced CPU usage

- Fixed noise, drop-out while changing Mode

- Dropped Zip file support

Available Formats

VST3 / AU / AAX Native - Audiosuite | Mac 10.15 or Higher | Intel / Apple Silicon

VST3 / AAX Native - Audiosuite | Windows 10 - 11

Mac Versions require a graphic card that supports “Metal”

Windows Versions require a graphic card that supports “OpenGL”

INSTALLATION INSTRUCTIONS FOR WINDOWS ZIP USERS

1. Put plug-ins into your plug-in folders.

2. Create a folder into 'C:\Users\Public\' and name it 'AnalogObsession'.

3. Download 'MoMa Content' and extract. There will be folder named 'MoMa'

4. Put 'MoMa' folder into 'AnalogObsession' folder.

The path should look like this;

'C:\Users\Public\AnalogObsession\MoMa'

PATREON[/CENTER]

[bitman]

Thanks a bunch for this. What does it "learn"?

[Burillo]

Thanks a bunch for this. What does it "learn"?

clearly it doesn't learn GUI size

[Erik_Lucas]

I hoped to see some more information on the machine learning aspect of this plugin.

[AnalogObsession]

I hoped to see some more information on the machine learning aspect of this plugin.

Hi,

Just replied similar request on GS and here is copy/paste.

It's something like other famous AI/ML plugins. Using Black Box method to achieve hardware's linear or non-linear response.

It uses FIR based algorithm and some other algorithms for learning and processing.

I wanted to release MoMa, first. It's based on DIY Mojo Maestro. Simple diode, resistor and capacitor circuit. I've already built it and using on my tracks while mixing and mastering.

Also, I captured my racked H*** Pro 790 channel strip. I'll release it, too. And more...

Thanks

[Erik_Lucas]

I hoped to see some more information on the machine learning aspect of this plugin.

Hi,

Just replied similar request on GS and here is copy/paste.

It's something like other famous AI/ML plugins. Using Black Box method to achieve hardware's linear or non-linear response.

It uses FIR based algorithm and some other algorithms for learning and processing.

I wanted to release MoMa, first. It's based on DIY Mojo Maestro. Simple diode, resistor and capacitor circuit. I've already built it and using on my tracks while mixing and mastering.

Also, I captured my racked H*** Pro 790 channel strip. I'll release it, too. And more...

Thanks

Interesting. I quite like the sound of MoMa. It's very heavy on my CPU though.

[Uncle E]

I hoped to see some more information on the machine learning aspect of this plugin.

I'm guessing it's like Tonex and NAM, which are also good for this kind of thing. The main problem with those is everyone thinks they're only for guitar amps. If Analog Obsession makes a plugin that captures and is marketed towards capturing studio devices, he'll have a real hit on his hands.

btw, the M-O mode sounds quite similar to the UA Studer A800 saturation.

[Tj Shredder]

Is the neural net capable to avoid aliasing by training it with alias free recordings??? That would be an advantage vs. classical modeling.
Would it be possible to use the neural processors of Apple Silicon processors?

[AnalogObsession]

Hello!

MoMa is just updated to Version 1.1.

Also, installation is changed. Please, delete plugins from your plugin folders before installing Version 1.1 and follow 'INSTALLATION INSTRUCTIONS'.

Now, MoMa uses shared folders for its content. So, you can use custom plug-in folders.

Please, check first post for instructions.

Thanks!

[AnalogObsession]

Is the neural net capable to avoid aliasing by training it with alias free recordings??? That would be an advantage vs. classical modeling.
Would it be possible to use the neural processors of Apple Silicon processors?

Hi,

I trained the unit at 96kHz and plug-in resampling internally. Aliasing is totally fine.

I'm using MBP M2 Max but it's hard to setup for GPU while training. So, I'm using Windows PC for training.

Thanks!

[AnalogObsession]

Good news!

Here is Version 1.2!

Now, installation process is fully automated!

Sorry for all the mess...

Hope, you can finally test it without any issue and like it!

Only Windows Zip users have to follow instructions.

Please, download again, install and enjoy!

[Uncle E]

It's a very good sounding plugin. Do you think you'll be able to reduce the high CPU usage in the future? I'm using a MacBook Pro i9 and the CPU meter in Luna goes up ~20% when I turn on MoMa.

btw, MOMA is aka Museum Of Modern Art.

[bmanic]

Is the neural net capable to avoid aliasing by training it with alias free recordings??? That would be an advantage vs. classical modeling.
Would it be possible to use the neural processors of Apple Silicon processors?

Hi,

I trained the unit at 96kHz and plug-in resampling internally. Aliasing is totally fine.

I'm using MBP M2 Max but it's hard to setup for GPU while training. So, I'm using Windows PC for training.

Thanks!

Actually you should probably check it again. It's aliasing A LOT.. if you run it at 44.1kHz projects it's crossing the line of being a bit too much aliasing for many tasks (we are talking -60dBFS peaks that mirror way back into the 500Hz range and even below).

So either the upsampling you are talking about is broken or then something else is.

Running this at 96kHz reduces aliasing a lot and it becomes very usable with aliasing down into the -90dBFS range at high frequencies and virtually zero below 5kHz.

[bmanic]

btw, the M-O mode sounds quite similar to the UA Studer A800 saturation.

The M-O and M-E mode here cuts pretty much all highs completely. Is it not doing that for you? Like there is a simple LPF filter from about 2kHz.

I hoped to see some more information on the machine learning aspect of this plugin.

I'm guessing it's like Tonex and NAM, which are also good for this kind of thing. The main problem with those is everyone thinks they're only for guitar amps. If Analog Obsession makes a plugin that captures and is marketed towards capturing studio devices, he'll have a real hit on his hands.

There are plenty of captures from other hardware, at least for NAM. Heck there are a few rather shady companies selling captures for NAM, trying to capitalize on this (shady in a way that they are probably not actually capturing owned hardware but rather other plugins and renting services like Access Online).

[imrae]

A deep neural net is, by definition, a big pile of nonlinear processes. There is going to be a significant CPU/anti-aliasing trade-off to implement these things.