AI audio-to-MIDI transcriber

Turn a recording into real MIDI notes using artificial intelligence — all processed in your own browser, nothing uploaded anywhere.

If HertzKit was useful to you, you can buy us a coffee. It's never required to use any tool.

How to use the transcriber

Upload a recording of an instrument or a voice singing a melody, up to 3 minutes long — that's a limit we chose so processing doesn't take forever on an average computer, not a restriction of the model itself. If you need to transcribe something longer, split it into parts.
Press “Transcribe”. The first time, it loads a ~900 KB AI model — after that it's cached by your browser.
The analysis runs entirely on your device: the audio is never uploaded to any server.
When it's done, you'll see a piano-roll of the detected notes and can download the result as a MIDI file, to open in any notation software or DAW.
Works best with one sound source at a time (one instrument or one voice) — it doesn't separate different instruments playing together, and it doesn't transcribe untuned percussion.

Uses Spotify's open-source Basic Pitch model (Apache 2.0 license). See full credits.