SeparateTracks / README.md
Surn's picture
Refactor audio processing to improve job ID handling and update .gitignore for separated files
18eff79
---
title: SeparateTracks
emoji: 🎼
colorFrom: red
colorTo: yellow
sdk: docker
sdk_version: 6.13.0
app_file: app.py
tags:
- audio
- music
- tools
- MCP
pinned: false
short_description: Separate tracks from mixed audio
---
Check out the configuration reference at <https://huggingface.co/docs/hub/spaces-config-reference>
## SeparateTracks
Download audio from YouTube, separate it into instrument stems with Demucs,
and view the results in a Gradio AudioGallery UI with MCP support.
## Features
- Accepts either a raw YouTube video ID or a full YouTube URL.
- Accepts `.wav` or `.mp3` uploads as a local audio override for separation.
- Builds a 7-track output gallery for drums, vocals, guitar, bass, other,
piano, and combined music, with a download link under each play button.
- Exposes an MCP endpoint at `/gradio_api/mcp/sse`.
## Usage
Run the app locally (not admin):
```bash
pip install -r requirements.txt
python app.py
```
```debugging
py -m app -Xfrozen_modules=off
```
Then open `http://localhost:7860` and use one of these flows:
1. Paste a YouTube video ID or full URL into `YouTube Video ID or URL` and
click `Separate Tracks`.
2. Upload a `.wav` or `.mp3` file into `Audio File Override (.wav or .mp3)`
and click `Separate Tracks`.
If an upload is present, it takes precedence over the YouTube field.
## Outputs
Separated files are written under `separated/htdemucs_6s/{job_id}/`.
For YouTube sources, `job_id` is the sanitized video title when available,
falling back to the video ID. For uploaded audio, `job_id` is a sanitized
version of the uploaded filename stem. Each stem card in the gallery can be
played in place or downloaded directly.
## Extractors & Cookies
For details about yt-dlp extractor behavior, see the official documentation:
- <https://github.com/yt-dlp/yt-dlp-wiki/blob/master/Extractors.md>
If you need authenticated extraction (for age-restricted or otherwise protected
content), one way to provide cookies safely is via a private/incognito browser
session:
1. Open a new private browsing/incognito window and log into YouTube.
2. In the same window and same tab from step 1, navigate to
<https://www.youtube.com/robots.txt> (this should be the only
private/incognito browsing tab open).
3. Export youtube.com cookies from the browser, then close the private/
incognito window so that the session is never opened in the browser again.
Upload the exported cookies file in the app UI when prompted.