Music_Source_Separator_Web_Application

Using trained models to separate music into its stems and available as a web application.

Application use Fast API and Demucs model

Getting Started

Prerequisites

• Python 3.13
• Homebrew (macOS)
• FFmpeg (version 7.x — required by torchcodec/torchaudio for audio decoding)

Install FFmpeg 7 via Homebrew:

brew install ffmpeg@7
brew link ffmpeg@7 --force

Setup

1. Clone the repository:

   git clone https://github.com/noeltoms/MSS_WebApp.git
   cd MSS_WebApp

2. Create and activate a virtual environment:

   python3 -m venv venv
   source venv/bin/activate

3. Install dependencies:

   pip install --upgrade pip
   pip install -r requirements.txt

Running the app

uvicorn src.main:app --reload

Then open http://127.0.0.1:8000 in your browser.

Upload an audio file, wait for processing to complete, and download the separated stems (vocals, drums, bass, other).

Notes

• The first run will download the Demucs model weights (~300MB+), which are cached locally afterward.
• Uploaded files and separated stems are stored temporarily in uploads/ and outputs/ — both are excluded from version control.