Audio to MIDI & Sheet Music
Upload a recording of an instrument or voice and get the notes from the audio back: a piano-roll view, a simplified staff notation, and a downloadable MIDI file. Transcription runs with Spotify's open-source Basic Pitch neural network, loaded and executed entirely in your browser via TensorFlow.js - nothing is uploaded anywhere.
Reviewed by Vincenzo Manto, Founder & Lead Developer · Last updated October 9, 2026
How it works
- 1
Upload an audio file - a melody, a chord progression, a vocal line, any instrument.
- 2
The browser downloads Basic Pitch's neural network model and downmixes/resamples your audio to the 22.05 kHz mono format it expects.
- 3
The model detects pitches, onsets, and note durations frame by frame, entirely on your device.
- 4
Review the detected notes as a piano roll and a simplified staff, then download them as a .mid file.
From recording to notes, without leaving the browser
Automatic music transcription usually means installing Python, wrangling model weights, or paying for a cloud API. This tool packages Spotify's Basic Pitch model as a static set of TensorFlow.js weights served alongside the page, so the entire pitch-detection pipeline - resampling, inference, note decoding - runs on your own machine.
Piano roll, staff, and MIDI from the same detection pass
A single transcription pass produces timed note events with pitch and amplitude. Those events feed three views at once: a time-accurate piano roll, a quantized staff notation for a quick visual read, and a hand-written Standard MIDI File encoder that turns the same notes into a .mid file you can drop into any DAW.
Frequently asked questions
Is my audio uploaded to a server?
No. The Basic Pitch model files are fetched once and run locally with TensorFlow.js in your browser tab; the audio itself never leaves your device.
What is Basic Pitch?
Basic Pitch is an open-source, instrument-agnostic polyphonic transcription model released by Spotify's Audio Intelligence Lab, published at ICASSP 2022. This tool uses the official TensorFlow.js build of that model.
Does it work with chords and multiple instruments at once?
Yes, Basic Pitch is polyphonic, so it can detect several simultaneous notes. Accuracy is best with one instrument or voice at a time; dense mixes with drums and vocals together are harder to transcribe cleanly.
Why is the sheet music simplified?
The staff notation snaps each detected note onto a 16th-note grid at a fixed tempo to render legible rhythm, since Basic Pitch outputs exact note timing rather than a quantized rhythmic value. For precise timing, use the downloaded MIDI file or the piano-roll view instead.
What can I do with the MIDI file?
Import it into any DAW (Ableton, Logic, FL Studio, GarageBand, Reaper) or notation software (MuseScore, Finale) to edit, re-score for a different instrument, or build arrangements from the transcription.