alppo/vae-conditioned-diffusion-model_v2

Variational Autoencoder Conditioned Diffusion Model

This model is designed to generate music tracks based on input playlists by extracting the "taste" from the playlists using a combination of a Variational Autoencoder (VAE) and a conditioned diffusion model.

Model Details

• —VAE: Learns a compressed latent space representation of the input data, specifically mel spectrogram images of audio samples.
• —Diffusion Model: Generates new data points by progressively refining random noise into meaningful data, conditioned on the VAE's latent space.