I got frustrated with existing DJ metadata tools acting as “black boxes”—uploading my audio to cloud servers and giving confident (but often wrong) key detection for minimal/tech-house tracks.
So I built Atomic Afterglow. It’s a Python-based analyzer that runs in the browser but processes audio locally (zero retention/no uploads).
How it works:
Engine: Built on Librosa and Streamlit.
Drop Detection: Instead of finding the “quietest” section, it uses a rolling RMS window to find the deepest structural energy trough in the middle 60% of the track (the breakdown/mix point).
Harmonics: Uses the Krumhansl-Schmuckler algorithm to generate the Top 3 Key candidates (Camelot formatted) with confidence margins, rather than forcing a single result on ambiguous tracks.
Repo (Open Source): https://github.com/phasesequencer/atomic-afterglow
I’d love to hear if the key detection holds up on your more complex or detuned tracks.
Comments URL: https://news.ycombinator.com/item?id=46906176
Points: 1
# Comments: 0