End-to-end Music Remastering System

This repository includes source code and pre-trained models of the work End-to-end Music Remastering System Using Self-supervised and Adversarial Training by Junghyun Koo, Seungryeol Paik, and Kyogu Lee.

We provide inference code of the proposed system, which targets to alter the mastering style of a song to desired reference track.

Pre-trained Models

Inference

To run the inference code,

1. Download pre-trained models above and place them under the folder named 'model_checkpoints' (default)
2. Prepare input and reference tracks under the folder named 'inference_samples' (default).
   Target files should be organized as follow:

    "path_to_data_directory"/"song_name_#1"/input.wav
    "path_to_data_directory"/"song_name_#1"/reference.wav
    ...
    "path_to_data_directory"/"song_name_#n"/input.wav
    "path_to_data_directory"/"song_name_#n"/reference.wav

3. Run 'inference.py'

python inference.py \
    --ckpt_dir "path_to_checkpoint_directory" \
    --data_dir_test "path_to_directory_containing_inference_samples"

4. Outputs will be stored under the folder 'inference_samples' (default)

Note: The system accepts WAV files of stereo-channeled, 44.1kHZ, and 16-bit rate. Target files shold be named "input.wav" and "reference.wav".

Configurations of each sub-networks

A detailed configuration of each sub-networks can also be found at

Self_Supervised_Music_Remastering_System/configs.yaml