A singer who describes themselves as bad but no longer terrible built OnKey, a free, open-source Mac app that helps people practice singing along to the songs they love. The developer grew frustrated when a singing app they used filled its song list with non-original versions, since they find it easier to improve by singing along to the original recording. After an AI agent was able to show their pitch in the terminal but could not detect the singer's pitch in real time, they built OnKey instead.
OnKey listens to a song once without saving it and learns the singer's pitch track, which takes a couple of minutes. Users can then sing along to that track. It works with Spotify, Apple Music, and YouTube playing in a browser. Lyrics come from LRCLIB, and song details are found through a short Shazam fingerprint. The app says it never sends the music or the user's voice anywhere, and that everything expected to stay private does.
The developer is candid about its limits. Noisy backing tracks make the pitch line hard to isolate, the app sometimes picks a note an octave off, and it only runs on Apple Silicon Macs with M1 or later and macOS 14.2 or newer. The scoring is also tuned for beginners, so it is fairly forgiving.
OnKey is released under the MIT license. The developer invites people to try it and report what works and what doesn't. A demo video is available on the project site, singonkey.app.
Source: Hacker News · Summarized by HeadlinesBriefing