Ad Audio Stem Split MVP

This MVP adds a local web app to split an ad audio clip into:

• Speech / Vocals
• Background Music / Noise

It includes:

• A FastAPI backend upload endpoint that runs audio_separator.separator.Separator
• A React + Vite frontend with stacked labeled waveforms and audio players

Folder Structure

• mvp/backend/app.py - API server
• mvp/backend/requirements.txt - backend web dependencies
• mvp/frontend/ - React UI

Run Backend

From the repository root:

python3 -m venv .venv
source .venv/bin/activate
python -m pip install -e "./audio-seperator[cpu]"
python -m pip install -r mvp/backend/requirements.txt
uvicorn mvp.backend.app:app --reload --port 8002

If your virtual environment already exists for this repository, just activate it and run the last two commands.

Backend endpoints:

• GET /api/health
• POST /api/separate (multipart form with file, optional model_filename)
• GET /outputs/... static output audio files

Run Frontend

In a second terminal:

cd mvp/frontend
npm install
npm run dev

Open the URL printed by Vite (typically http://localhost:5173).

Notes

• Supported input formats include: .wav, .mp3, .flac, .m4a, .mp4, .ogg, .opus, .aiff, .ac3.
• The first separation request may take longer because model files are downloaded and the model is loaded.
• The backend caches loaded model instances by model_filename to make later requests faster.
• The MVP default model is UVR_MDXNET_Main.onnx for faster turnaround in ad splitting workflows.
• The frontend model picker includes a Super Fast option (UVR_MDXNET_KARA_2.onnx) for quickest iteration.