Re-Beethoven

An attempt to regenerate Beethoven's works using deep learning.

experiment

Loss: AuraLoss, GAN Wrapper(EncodecDiscriminator in stable audio tools)

Model (Uncond): Encoder + DiT(Based on Differential Transformer) + Decoder

Autoencoder = Encoder + BottleNeck(VAE,AE) + Decoder
    trained before and freezed for Diffusion

Dataset: Train: Musicnet(Kaggle)