MSST-WebUI

English | 简体中文 | 繁體中文 | 日本語 | 한국어
A WebUI app for Music-Source-Separation-Training and we packed UVR together!

Introduction

This is a webUI for Music-Source-Separation-Training (MSST), which is a repository for training models for music source separation. You can use this webUI to infer the MSST model and VR Models, and the preset process page allows you to customize the processing flow yourself. You can install models in the "Install Models" interface. If you have downloaded Ultimate Vocal Remover (UVR) before, you do not need to download VR Models again. You can go to the "Settings" page and directly select your UVR5 model folder. We also provide some convenient tools in the WebUI such as Singing-Oriented MIDI Extractor (SOME), advanced ensemble mode, and more.

Usage

• Windows: Download the installer from Releases or clone this repository and run from source.
  
• Linux/macOS: Clone this repository and run from source.
  
• Google Colab: Click here to run the webUI on Google Colab.

Download links

Websites	Download Links	Notes
Github Releases	https://github.com/SUC-DriverOld/MSST-WebUI/releases	Only installer, no models
Huggingface	https://huggingface.co/Sucial/MSST-WebUI/tree/main	Installer and all available models

Documents

We provided some detailed chinese documents for chinese users, click here to jump. For other users, go to docs folder to find some documents. You can also see deton24's Instrumental and vocal & stems separation & mastering guide, which is a great guide too.

Run from source

• Clone this repository.

  git clone https://github.com/SUC-DriverOld/MSST-WebUI.git
  cd MSST-WebUI

• Create Python environment and install the requirements. We recommend to use python 3.10.

  conda create -n msst python=3.10 -y
  conda activate msst
  pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
  pip install -r requirements.txt --only-binary=samplerate

• After installing the requirements, go to site-packages folder, open librosa\util\utils.py and go to line 2185. Change the line from np.dtype(complex): np.dtype(np.float).type, to np.dtype(complex): np.dtype(float).type,. If you do not know how to do this, you can use the following command.

  pip uninstall librosa -y
  pip install tools/webUI_for_clouds/librosa-0.9.2-py3-none-any.whl

• Run the webUI use the following command.

  python webUI.py

• For more detailed launching arguments, or if you want to run the webUI on a cloud platform, see this document.

• You may meet the following error when using model_type swin_upernet: ValueError: Make sure that the channel dimension of the pixel values match with the one set in the configuration. Refer to this issue to solve the problem.

CLI & Python API

See this document for more details.

Training & Validation

See this document for more details.

Reference

• [deton24's Documents] Instrumental and vocal & stems separation & mastering guide
• [KitsuneX07's ComfyMSS] ComfyMSS
• [PyQt-Fluent-Widgets] PyQt-Fluent-Widgets
• [Pyside6] PySide6 documents
• [python-audio-separator] python-audio-separator
• [Singing-Oriented MIDI Extractor] SOME
• [Ultimate Vocal Remover] Ultimate Vocal Remover
• [ZFTurbo's MSST code] Music-Source-Separation-Training

Thanks to all contributors for their efforts

Thanks for all model providers

anvuew	aufr33	deton24	jarredou	pcunwa	Sucial
Super-YH	viperx	wesleyr36	yxlllc	ZFturbo