U-Net_MUSDB18

デモページ(demo)

Implementation of U-Net for music separation.

MUSDB18 dataset is used for training and evaluation. You can download it here (access request is necessary).

Prerequisites

• librosa >= 0.7.0
• numpy >= 1.16.0
• pysoundfile >= 0.10.0
• pytorch >= 1.2.0
• tensorboard >= 1.14.0
• torchaudio >= 0.3.0

and ffmpeg should be installed to do the preprocessing.

Usage

The original MUSDB18 dataset consists of .mp4 files. First, convert the whole dataset into .wav format for convenience.

$ python convert_to_wav.py --sr 22050 {your_path}/musdb18 {your_path}/musdb18_wav_22050

Then train the model by train.py. For example,

$ python train.py --dataset {your_path}/musdb18_wav_22050 \
                  --gpu 0 1 2 3 \
                  --batch-size 256 \
                  --output ./model

Help message of commandline arguments can be found by

$ python train.py -h

You can use separate.py to test your model.