Audio to MIDI

What it does

Audio to MIDI estimates a pitch contour from a monophonic source — one note at a time — and writes a minimal Type-0 MIDI file you can open in a DAW.

Polyphonic material (chords, full bands) cannot be separated reliably without machine-learning models. For best results use a solo voice, bass line, or single synth lead with limited reverb.

Pitch is tracked with autocorrelation on short frames; notes are quantized to a simple grid for readability.

Why monophonic-only

Autocorrelation finds the strongest repeating period in a waveform, which works well when one note sounds at a time. When several notes overlap — a guitar chord, a full mix — the algorithm has to guess which periodicity matters, and it guesses wrong often enough that the output becomes unusable. Dedicated polyphonic transcription needs a trained model to separate simultaneous notes, which is outside the scope of a lightweight browser tool.

Getting a cleaner transcription

Isolate the melody first: a vocal take, a bass DI, a whistled or hummed line, or a solo instrument recording all track well. Heavy reverb, vibrato, and pitch bends can confuse frame-by-frame tracking, so a drier take with a stable pitch will need less manual correction once it lands in your DAW.

Isolate a clean section first with Audio Trimmer, or confirm the key of the result with the Key Detector.

How to use it

Tip: Hummed or whistled melodies often transcribe more cleanly than recorded instruments, since they have a simple, stable pitch with little harmonic interference.