Audio to MIDI

It runs Spotify’s Basic Pitch, a neural network (TensorFlow.js) model, entirely in your browser. The model listens for note onsets and pitches, then produces MIDI note events with timing, pitch, and velocity. No audio is ever uploaded.

Monophonic and simple polyphonic content - solo piano, guitar, vocals, flute, or a single lead line - transcribes most accurately. Dense mixes, heavy distortion, and drums produce noisier results because pitch is ambiguous.

The local model assets load once, then inference runs on your device. Processing time depends on the recording length and available CPU, so older phones and dense polyphonic audio can take longer. Only the first 120 seconds are analyzed to keep the browser workload bounded.

Minimum confidence filters out weak or uncertain note detections - raise it to keep only strong notes. Minimum note length removes very short blips. Both re-derive instantly from the cached analysis without re-running the model.

Yes. The download is a Standard MIDI File (format 0) with tempo and note on/off events. Import it into Ableton, Logic, FL Studio, GarageBand, MuseScore, or any DAW to edit, re-instrument, or notate.

Multiple instruments, drums, effects, and overlapping notes make individual pitches ambiguous. Solo melodies, piano lines, guitar parts, and vocals usually produce cleaner MIDI.