ClearStage is an AI-powered audio engine designed to clean up and enhance live concert recordings.
The goal of this project is to take a raw live audio recording and intelligently separate the stems (vocals, drums, bass, other) to reduce crowd noise and improve vocal clarity, resulting in a studio-quality mix.
I started this project to assist in Roehampton's Cabaret Night live recordings which were of poor quality.
- AI Source Separation: Uses Demucs (HTDemucs_FT) for high-fidelity stem separation.
- Hard Center Vocal Extraction: Removes stereo bleed using Mid-Side processing to isolate vocals.
- Crowd Taming: Smooths stereo width and removes harsh transients from the crowd noise.
- Smart Vocal Ducking: Automatically lowers backing tracks when vocals are present.
- Vocal Polishing: Applies parallel compression and dynamic expansion for studio presence.
- Auto-Mastering: Recombines stems with a "ClearStage" mixing profile.
python clearstage_engine.py input_file.mp3