Apppublic

The8dEngine/Stems

sourceHugging Faceupdated 3mo agoView on Hugging Face

0likes

App README

8D Engine — Demucs stem separation

A tiny free-tier (CPU, 16 GB RAM) Hugging Face Space that separates a track into vocals / drums / bass / other with Demucs, so the 8D Engine app can offer downloadable stems without running PyTorch on its own small server.

The 8D app calls the named API endpoint `/separate` via gradio_client:

python

from gradio_client import Client, handle_file
client = Client("YOUR_USERNAME/8d-demucs-stems")
vocals, drums, bass, other = client.predict(handle_file("song.mp3"), api_name="/separate")

Free CPU Spaces sleep when idle; the first request after a nap has a cold-start delay (~1–2 min) while the model loads, then subsequent requests are faster.