Sound generation using VAE, conditioned on musicnn tags.
You can download a tiny model here.
Find some sample melspecs + extracted tags are in the data directory, to play with the Generate notebook without training again.
Find some generated samples in the generations dir.
! A coolest CNN version coming soon (I hope) !
If you see some inconsistencies or errors, feel free to reach out and ask for a pull request, I am human after all :)
🎧 If you use it for your musical projects, send some links over, I am always listening to new music!
- Dataset: generate your own dataset.
- Train: train the model, playground for generation, latent space visualization.
- Generate: playground for generation + an attempt of style transfer to condition the generation having one tag to match another one.
- https://github.com/jordipons/musicnn
- Took a lot of inspiration from https://github.com/s-omranpour/X-VAE-keras for the network architecture.
