Do my files get uploaded to your servers?

No. Your browser reads the file locally and runs Spotify Basic Pitch inside this tab. After the page paints, it downloads a 1 MB model from this site; that request is our static asset, not your audio. This tool has no processing endpoint, so there is nowhere for an upload to go.

Is there a cost, a sign-up or a watermark?

None of the three. Transcription never touches your credits, and the output is a normal Standard MIDI File with no extra tracks or watermark.

Will this work on a mixed song with drums and vocals?

Poorly. The model is built for one instrument — a guitar take, a piano sketch, a hummed melody. A full mix smears competing pitches into extra notes. Isolate the part you want, then transcribe that file.

What model is this, and how accurate is the MIDI?

Spotify Basic Pitch (Apache-2.0, 2022), running in TensorFlow.js. It is polyphonic and instrument-agnostic. Notes are quantized to 12-tone equal temperament; we do not write pitch-bend data, because the model's bend values are not MIDI pitch-wheel. Treat the file as a sketch to edit, not a finished score.

Audio to MIDI

Drop a recording of a melody, guitar or piano. This tab transcribes the notes and gives you a Standard MIDI File you can open in any DAW. Mixed songs with drums and vocals come out messy — isolate one instrument first.

Runs on your deviceFree, no creditsNo sign-up

Your audio stays on this device The file is decoded and transcribed in this tab. Music3AI never receives it. The only extra download is a 1 MB transcription model from this site, and that request does not include your audio.

Drop an audio file here

MP3, WAV, M4A, OGG, FLAC. One instrument, or a hummed melody, works best.

Up to 40 MB or 3 minutes per file. Several files transcribe one after another.

Queue (0)

Nothing queued yet. Add a file above and it will appear here.

How to turn audio into MIDI

01
Drop a recording of one instrument or a hummed melody (MP3, WAV, M4A, OGG or FLAC), up to 3 minutes.
02
Press Transcribe. The model runs in this tab — a 30-second clip usually finishes in under a minute on a laptop.
03
Download the .mid file and open it in a DAW, a notation app or a MIDI player.

Audio to MIDI questions

Do my files get uploaded to your servers?: No. Your browser reads the file locally and runs Spotify Basic Pitch inside this tab. After the page paints, it downloads a 1 MB model from this site; that request is our static asset, not your audio. This tool has no processing endpoint, so there is nowhere for an upload to go.
Is there a cost, a sign-up or a watermark?: None of the three. Transcription never touches your credits, and the output is a normal Standard MIDI File with no extra tracks or watermark.
Will this work on a mixed song with drums and vocals?: Poorly. The model is built for one instrument — a guitar take, a piano sketch, a hummed melody. A full mix smears competing pitches into extra notes. Isolate the part you want, then transcribe that file.
What model is this, and how accurate is the MIDI?: Spotify Basic Pitch (Apache-2.0, 2022), running in TensorFlow.js. It is polyphonic and instrument-agnostic. Notes are quantized to 12-tone equal temperament; we do not write pitch-bend data, because the model's bend values are not MIDI pitch-wheel. Treat the file as a sketch to edit, not a finished score.

Want the song instead of the MIDI?

Music3AI turns one sentence into a finished track with vocals, lyrics and a mix. 32 styles, up to 5 minutes, MP3 download.

Make my first song