Before and after

Examples

Three original clips run through MidiDraft, with the test conditions and what to expect from each — including one the model handles badly.

Every clip here was synthesised from scratch for this site, so there are no third-party rights on them and you are free to re-run the test yourself. Nothing on this page is a claimed accuracy figure — open a clip in the converter and judge the result with your own ears.

Clear single-note piano melody

One note at a time, moderate tempo, no effects. This is the material audio-to-MIDI handles best.

What to expect
Expect nearly every note start and pitch to come through. You may still want to trim note ends before dropping it into a project.
Test conditions
Synthesised piano-like tone, 96 BPM, 22 seconds, mono WAV at 44.1 kHz, no reverb or compression.
Convert this in the browserDownload the source audio

Plucked guitar arpeggio

Four chord shapes played one string at a time, with notes ringing into each other.

What to expect
Pitches land, but overlapping ring-out produces longer notes than you played, and sometimes a doubled onset. "Trim overlaps" and "merge repeated notes" are the two controls that help here.
Test conditions
Synthesised plucked tone, 96 BPM, 18 seconds, mono WAV at 44.1 kHz, natural ring-out, no effects.
Convert this in the browserDownload the source audio

Dense chords with reverb — where this breaks down

Hard case

Five-note pad chords, a melody on top, added reverb and a noise floor. Closer to a finished mix than to a clean take.

What to expect
Expect a rough sketch at best: missed inner voices, extra notes smeared by the reverb tail, and an unreliable melody line. Use this to judge whether the tool suits your material before you pay for anything.
Test conditions
Synthesised pad chords plus melody, 96 BPM, 16 seconds, mono WAV at 44.1 kHz, artificial reverb at roughly 45% wet and a low noise floor.
Convert this in the browserDownload the source audio

Why there is no accuracy percentage

A single number hides the thing that matters: which material works. Three clips are not a benchmark either — they are a way to hear the shape of the output before you spend time on it. If you want to know whether your own material converts, the honest test is to run thirty seconds of it through the free converter, which costs you nothing and needs no account.

What makes a take convert well