Prerequisites

# Debian 13 / Ubuntu 25.04
sudo apt install -y pyenv

cat <<EOF >> ~/.bashrc

export PYENV_ROOT="\$HOME/.pyenv"
eval "\$(pyenv init -)"
EOF

source ~/.bashrc

pyenv install 3.11.9

sudo apt install -y pipenv ffmpeg
pipenv install --verbose

GPU access (Radeon VII / gfx906)

# Check if your part of the render group
groups

# Add yourself to the render group
sudo usermod -a -G render $USER
newgrp render

Usage

Two-pass pipeline for cleanest results:

pipenv shell

# Pass 1: best-quality vocal / instrumental split
audio-separator ./still_be_there.mp3 -m vocals_mel_band_roformer.ckpt --output_dir ./stems

# Pass 2: break the INSTRUMENTAL into instruments
audio-separator "./stems/still_be_there_(other)_vocals_mel_band_roformer.flac" -m htdemucs_6s.yaml --output_dir ./stems/instruments_6s/

# Fallback:
# audio-separator ./still_be_there.mp3 -m htdemucs_ft.yaml --output_dir ./stems

# Reset the GPU without rebooting
# sudo modprobe -r amdgpu && sudo modprobe amdgpu

# scp -r ./stems/ dste@192.168.1.188:/workspace/