d3net-spleeterweb 0.1.3

Unofficial Python package of D3Net implementation by Sony Research AI, used in Spleeter Web.

D3Net (Music Source Separation) for Spleeter Web

This is a modified version of the official D3Net repo made to be compatible with Spleeter Web!

This is inference code for D3Net based music source separation.

Quick Music Source Separation Demo by D3Net

From the Colab link below, you can try using D3Net to generate and listen to separated audio sources of your audio music file. Please give it a try!

Getting started

Prerequisites

• nnabla
• librosa
• pydub
• numpy
• soundfile
• yaml

Inference: Music source separation with pretrained model

Download the pre-trained D3Net model for Music Source Separation here.

Run the below inference command for a sample audio file test.wav in current directory:

 python ./separate.py -i ./test.wav -o output/ -m d3net-mss.h5 -c cudnn

Arguments:
-i : Input files. (Any audio format files supported by FFMPEG.)
-o : Output directory. (Output folder path to save separated instruments)
-m : Model file. (Pre-trained model)
-c : Context. (Extension modules : cpu or cudnn)

Training: Train the music source separation model from scratch (coming soon)