Does it work on rave v.2 model?

[gyuchulm]

I tried to build my own model with v.2 and there are some issues with the build model, and trying to use it in Supercollider.

I can run the pre-trained model in my SC, but v2 model that I trained are not working in the example code.

I saw your post in Rave GitHub, did your model works in this code?

[gyuchulm]

And also, a bit complex is the latent size, is this a buffer size in a training stage? I couldn't find the options of latent size in a rave code.

I trained the model with rave v.2, and the sound quality is a bit weird, but I can hear the sound. But the supercollider has no sounds.

[victor-shepardson]

It should be fixed now, apparently RAVE used to export sampling_rate but in v2 no longer does, which was crashing rave-supercollider.

[victor-shepardson]

And also, a bit complex is the latent size, is this a buffer size in a training stage? I couldn't find the options of latent size in a rave code.

the number of useful latents is determined by training (see the fidelity_0.99 curve in tensorboard), and the exact number you see in nn~ or rave-supercollider is set when you use rave export with the --fidelity argument