Audio Stem Separator 🎵

Ever wanted to isolate just the vocals from your favorite song? Or maybe you're a producer who needs to pull out that perfect drum track? This tool uses AI to magically separate audio into individual parts (we call them "stems") - vocals, drums, bass, piano, and more!

See It In Action

Web Version

Just drag, drop, and let the magic happen

Watch your audio get separated in real-time

Download exactly what you need

Desktop App

Clean, modern interface that's actually fun to use

No more wondering "is it done yet?"

Your stems, ready to download

What Makes This Cool?

Smart AI Separation - We're talking studio-quality results that'll make you wonder how it works (spoiler: lots of math and neural networks)

Super Easy to Use - Literally just drag your file in. That's it. No PhD in audio engineering required.

Watch It Work - See your audio get processed in real-time because waiting without knowing what's happening is the worst

Get What You Want - Download individual parts or grab everything at once

Plays Nice - Works with pretty much any audio format you throw at it (MP3, WAV, FLAC, you name it)

Smart Cleanup - Automatically tidies up so your hard drive doesn't hate you

Getting Started (The Fun Part!)

What You'll Need

• Python 3.8 or newer (if you don't have it, grab it from python.org)
• A graphics card with CUDA is nice but not required (it just makes things faster)

Setting Things Up

Step 1: Get the code

git clone 
cd audio-stem-separator

Step 2: Create your own little Python world

python -m venv venv
source venv/bin/activate  # Windows folks: venv\Scripts\activate

Step 3: Install the magic

pip install -r requirements.txt

Fire It Up!

python -m app.main

Then head over to http://localhost:8000 and start separating some audio!

Actually Using This Thing

For Tinkering Around

Start the server:

uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

Then visit:

• http://localhost:8000 for the main app
• http://localhost:8000/docs if you want to see all the nerdy API details

For Serious Business

Production mode (handles more traffic):

uvicorn app.main:app --host 0.0.0.0 --port 8000 --workers 4

Or if you're into Docker:

docker build -t audio-stem-separator .
docker run -p 8000:8000 audio-stem-separator

Tweaking Settings

You can customize how things work with these environment variables:

What	What It Does	Default
MODEL_PATH	Where to find the AI models	models/
UPLOAD_DIR	Where uploaded files go	uploads/
MAX_FILE_SIZE	Biggest file you can upload	100MB
GPU_ENABLED	Use your graphics card for speed	true

The Desktop App Experience

Want something that feels more like a real program? We've got you covered with a desktop version that's actually pleasant to use.

What You Need

• Python 3.8+ (same as before)
• PyQt6 (for the pretty interface)
• The main server needs to be running

Getting It Running

Install the desktop bits:

pip install PyQt6 requests

Start the server first:

python -m app.main

Then launch the desktop app:

python desktop.py

The desktop app gives you drag-and-drop uploading, real-time progress updates, and easy downloading. It's like the web version but feels more like home.

Just Want the Web Interface?

If you prefer browsers, serve up the web interface like this:

python -m http.server 3000

Then visit http://localhost:3000 and you're good to go!

For the API Nerds

If you like diving into technical details, check out:

• Interactive docs: http://localhost:8000/docs
• Alternative docs: http://localhost:8000/redoc

What the API Can Do

How	Where	What
GET	/	"Hey, are you working?"
POST	/upload	"Here's my audio, do your thing"
GET	/status/{task_id}	"How's my audio coming along?"
GET	/download/{task_id}	"Give me all the stems!"
GET	/download/{task_id}/{stem_name}	"Just the vocals, please"
DELETE	/cleanup/{task_id}	"Clean up after yourself"
GET	/tasks	"Show me everything you're working on"
GET	/queue/status	"How busy are you right now?"

What's Under the Hood

We built this with some really solid tools:

FastAPI powers the backend (it's fast and handles everything gracefully)

Custom ML models do the actual audio separation magic

Redis keeps track of the queue (optional but helpful)

Local storage keeps your files safe

CUDA support makes GPU owners very happy

Want to Help Make This Better?

We love when people contribute! Check out our Contributing Guide to see how you can help make this even more awesome.

The Legal Stuff

MIT License - basically, use it however you want, just don't blame us if something goes wrong. Full details in the LICENSE file.