I could read 10 white papers and maybe figure it out.
It seems RAVE makes new waveforms out of other waveforms. Which is what Im interested in with my eII library. I think, Im not sure.
Flucoma seems more into slicing and dicing and morphing?
Just linking the homepages doesn’t really help, I get the idea, but they seem so similar. . A layman’s answer is much better. Maybe ill just ask ChatGpt
I asked Claude, I was close
I’ll provide a concise comparison of RAVE and FluidCorpusManipulation, focusing on their main differences, strengths, and approaches to sound generation:
RAVE (Realtime Audio Variational autoEncoder):
- Purpose: Real-time audio synthesis and manipulation
- Technique: Uses variational autoencoders (VAEs) for learning latent representations of audio
- Strengths:
- Fast, real-time operation
- Capable of generating high-quality audio
- Can interpolate between different sounds smoothly
- Best for:
- Live performance and real-time sound design
- Creating new hybrid sounds by blending existing ones
- Exploring latent space of audio for novel sound generation
FluidCorpusManipulation:
- Purpose: Granular synthesis and audio corpus manipulation
- Technique: Uses machine learning for descriptor-based granular synthesis
- Strengths:
- Precise control over audio characteristics
- Can work with large audio datasets (corpora)
- Offers more traditional granular synthesis techniques alongside ML approaches
- Best for:
- Detailed sound design and composition
- Working with large collections of audio samples
- Creating textures and evolving soundscapes
Main differences:
- Approach: RAVE uses deep learning to create a continuous latent space, while FluidCorpusManipulation focuses on granular techniques and descriptor-based synthesis.
- Real-time capability: RAVE is designed for real-time operation, while FluidCorpusManipulation may be better suited for offline processing and composition.
- Control: RAVE offers high-level control through latent space manipulation, while FluidCorpusManipulation provides more precise control over individual grains and audio descriptors.
- Sound sources: RAVE typically works with individual sounds or small datasets, while FluidCorpusManipulation is designed to handle large audio corpora.
Both tools are powerful for creating new sounds, but they excel in different areas. RAVE is better for real-time, intuitive sound blending and exploration, while FluidCorpusManipulation offers more detailed control and is well-suited for working with large audio datasets in composition and sound design contexts.
Would you like me to elaborate on any specific aspect of these tools?
1 Like
I noticed there is one build for Mac that is 32 megs and another that is 133 megs.
The 32 meg one seems to be doing stuff. I just hope I have the right thing installed in Extensions. Seems to be working.
There are actually three mac builds:
nn.ar-macOS-latest-arm64.zip: for M1/2 chips, build for latest macOS version (14 at the moment)nn.ar-macOS-12-arm64.zip: for M1/2 chips, build for macOS version 12nn.ar-macOS-x64.zip: legacy build for Intel macs (also macOS 12)
arm64 builds are lighter, intel is heavier. You should be fine with either of the lighter ones.
1 Like
While Claude’s summary is impressive, it is misleading in so many ways. I would take it with a grain of salt. Or maybe a few dashes.
Luckily, you don’t have to read 10 white papers. You could watch @tedmoore’s excellent video tutorials and the Rave tutorials on YouTube.
Honestly, having worked with this stuff for the past 5-7 years, there is no way around learning new things and learning to think in completely new ways. While getting something to work is relatively easy, getting the systems to do something you want to listen to takes work.
Sam
2 Likes
Right I see what you mean, I’ll look more into fluid
For what it’s worth, Claude can make very functional code, stuff that works, I’m only limited by having no subscription. You have to know some things about sc in order to point it in the right direction, keep feeding it errors, suggest stuff, and you have something I could never code. Like those keyboard patches I posted. GUIs are hard for me to code. I find them the most difficult. That’s how I’m using it mainly, but I have asked it to help me with some regular sc code. It might not understand rave or fluid, but it’s not too bad on sc, and this is early days,
Thank you , ll try these and report
Sorry I had missed your message! I haven’t tried yet with anything else than rave and msprior… but when I started writing nn.ar I had the same idea about it being able to load any torchscript
I’ll get back to you, but the code you wrote (using nn_tilde.Module) looks at least in the right direction to me, as basically we want to imitate whatever rave does at “export” time… have you tried it already?
First of all, your NN.ar is a lovely piece of engineering. Bravo for that.
I got a version working that will load a more standard model. Here are two regressor networks trained on a variable oscillator, frequency modulating one another (probably not a great use case considering the cpu use, haha):
Because of some of the features of the nn~ project (mainly the 3D tensors necessary for batch processing), I had to hack around quite a bit, so this should probably be a separate UGen. Maybe you would be able to make a universal one, but I think this would complicate the python side too much and make it mostly unusable.
I’ll clean this up and get it up on github. Hopefully you are cool with that.
BTW - dealing with tensors in c++ is a nightmare
Sam
1 Like
Can you explain what this means? What is a regressor in this instance? What is happening? Or is there a place to learn about this ? Thanks
See this video by Dr Moore:
and then watch this one (edit - here is the SC one):
Sam
3 Likes
Thank you. Soooo interesting.
How did you get that violin sample going? The Rave vst ? Or supercollider code?
The docs on Rave in SC are sparse. Are there any examples anywhere ? Example code?
Also, I tried your code.
RAVEEncoder doesn’t exist in the Rave I installed in SC for Mac.
Any resource online on using rave in sc ? Any example code. Maybe people using it can post what they’ve done?
I downloaded the Vst, and app. It seems to be geared to playing live instruments into it.
Hi @sslew !
Sorry for my late response!
For this example, I used Neutone models. You need to install Neutone and then find the one for the violin.
The models are located here: “/Users/jose/Library/Application Support/Qosmo/Neutone/models/2de8b5fdc596f834e1e2cbf89c56a0e5”.
I haven’t tested NN.ar yet, but I think it would work as well!
If you want, I can send you the entire example later.
Hope this helps!
José
1 Like
I’m using the vst plug in for fun, but I’m focusing on FluCoMa for supercollider. No need to send anything , thank you though
1 Like