OnKey is a free, open-source macOS singing trainer that listens to music you play in Spotify, Apple Music, or YouTube and shows where your voice is off-key. On first play it extracts lyrics and learns the lead vocal, then converts the melody into a per-syllable note lane; on subsequent plays it locates the song by listening and displays scrolling note bars with your live pitch overlaid, scoring each line and producing a per-line report. It separates vocals, tracks pitch with CREPE, uses a GAME note model to turn pitch into bars, and is forgiving of octave errors and slight timing delays. Requirements are Apple Silicon (M1+) and macOS 14.2+, and user setup grants microphone and system audio access.
Privacy and distribution are central: audio analysis happens in memory and learned data stays on your Mac, though optional network calls fetch song IDs, lyrics, covers, and update checks (Shazam, LRCLIB, iTunes search, GitHub). It doesn’t stream or save music, and sharing is limited to short videos without music. The project is MIT-licensed with a no-reuse-on-name/icon clause, credits models like wav2vec 2.0 and UVR MDX-Net, and can be built from source with Xcode tools, Rust, Node, and Python; custom builds lack the developer’s Shazam signature. Support, diagnostics, contribution instructions, and legal contact details are provided.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.