Deep-Classical-Music

Overview

This project focuses on exploring methods to generate Indian Classical music without lossing its original attributes. Currently we have explored unconditional and conditional generation using Jukebox methodoloy. This repo contains the training code for the same. In future, we will add more methods because with the dawn of Generative AI, new methods for audio generation are researched and implemented day by day.

TODO

• Upload the colab notebook for inference of Jukebox model
• Upload relevant links to dataset and checkpoints
• Upload the colab notebook for training of VQ-VAE and Upsamplers which make up the Jukebox model
• Make the code modular and train on more hardware
• Explore other methods for audio generation

Setup

Dataset

Flute Dataset

This is the unlabelled flute and tabla dataset collected from youtube majorly from the artist Hariprasad Chaurasia.
Run the following command to download the dataset:

!gdown --id 1qWBDK_SoMo471OycIwntjg1Qxx0fbgsu
!unzip flute_dataset.zip
!rm flute_dataset.zip

Pretrained Checkpoints

It is advised to store checkpoints in the following structure for easier access and readability.

content 
| - checkpoints 
  | - vqvae
  | - prior

VQ-VAE

Run the following command:

!gdown --id 1WQc4vOJXilI7gnn0iyKREe7AVcSKKpP1

Prior

Run the following command:

!gdown --id 1NahRriZJIPtZsFtFN9vOakVzgX5_mbzu

Inference

Uploaded inference.ipynb for generating samples in two ways :

• Ancestral Sampling
• Primed Sampling

Here is the live version of the notebook :

Credits

Jukebox - A Generative Model for Music

Support

There are many ways to support a project - starring⭐️ the GitHub repo is just one.