Audio to MIDI
Sometimes you hum a melody, or play a line on guitar, and what you actually want is the notes — so you can hand them to a synth, double them with a pad, print them as a bassline, or just see them written out. To MIDI listens to an audio clip, works out the melody, and drops it onto a MIDI track as editable notes.
It all happens on your device. Nothing is uploaded, no account is needed, and it works with no connection at all.
🔒 Pro feature. Audio to MIDI is available with Homecrate Pro. Free users see the To MIDI button with a lock; tapping it opens the upgrade screen.
Converting a Clip
- Make sure your project has at least one MIDI track. If it doesn’t, you’ll get a “No MIDI tracks” message — add one first. See MIDI Tracks Overview.
- Long-press the audio clip (or tap its ✏️ pencil icon) to open the Clip Edit modal.
- Tap To MIDI.
[SCREENSHOT: Clip Edit modal — To MIDI button in the action row]
The button shows a spinner while it analyses. A few seconds later the timeline switches to MIDI mode and your new clip is sitting there.
What You Get
| Detail | Behavior |
|---|---|
| Where it lands | On the MIDI track with the fewest clips, at the same timeline position as the source audio clip |
| Name | The source clip’s name with (MIDI) appended |
| Length | Rounded up to a whole number of bars at your project tempo, so no detected note ever falls outside the clip |
| Velocity | Derived from how loud each note was — quiet phrases come in softer |
| Instrument | None is assigned. The clip just appears; you route it to an instrument yourself |
| Timing | Follows the clip’s time-stretch, so a stretched source converts in time with what you hear |
Only the trimmed span of the clip is analysed — whatever is between the IN and OUT points, using the saved trim, not unsaved edits in the editor. Trim first, tap Apply, then convert.
Once it’s there it’s a completely ordinary MIDI clip. Open it in the piano roll to fix notes, quantize, transpose, or rewrite it entirely.
[SCREENSHOT: A vocal audio clip and the MIDI clip generated from it, aligned on the timeline]
What It’s Good At
The detector is monophonic — it tracks one pitch at a time, the same way a tuner does. That makes it strong on:
- Solo vocals — sung or hummed lines, the intended use case
- Isolated single-note instrument parts — bass lines, lead lines, whistled melodies, a clean guitar playing one note at a time
- Anything you’d play into a tuner and get a stable reading from
It’s built to be forgiving of real performances: vibrato is tolerated rather than chopped into a dozen notes, a brief dropout mid-phrase gets bridged instead of splitting the note, and an octave misread on a single frame gets rejected rather than printing a wrong note.
A sustained pitch move of about a semitone or more is read as a new note — that’s how slides and hammer-ons come through as separate notes rather than one smeared one.
What It’s Not Good At
Be realistic about this one; the limits are structural, not a tuning issue.
- Chords and polyphony. A strummed guitar, a piano chord, two singers — the detector can only report one pitch, so it will pick something and it will not be the chord. There’s no polyphonic mode.
- Drums and unpitched percussion. No stable pitch means nothing useful comes out. Use the drum sequencer instead.
- Full mixes. A stereo bounce of a whole song has everything happening at once. Expect noise.
- Heavy distortion and dense harmonics. Distorted guitar in particular confuses pitch detection badly.
- Very short notes. Anything under roughly 70 ms is dropped as noise. Fast trills and grace notes will thin out.
- Bleed and heavy reverb. A vocal recorded in a loud room, or drenched in reverb, gives the detector two pitches to argue about.
If nothing usable is found you’ll see “No melody detected — could not find pitched monophonic content in this clip.” That’s not a failure to fix; it’s the detector telling you the material isn’t monophonic and pitched.
The output is a starting point, not a transcription. Expect to spend a minute in the piano roll cleaning up octave jumps at phrase ends, trimming stray short notes, and quantizing. That’s normal, and it’s still far faster than playing the line in by hand.
Tips
- Isolate before you convert. If the part you want is buried in a mix, separate it out first — a clean, single-source clip is the single biggest quality factor.
- Trim to one phrase. Converting four bars of a clean vocal gives a far better result than converting a whole three-minute take, and it’s much quicker to tidy up.
- Sing it if you can’t play it. Humming a bassline into the mic and converting is often the fastest route to a part you couldn’t perform on a keyboard.
- Don’t fight a bad conversion. If the first attempt is a mess, trim tighter or clean the source rather than repairing dozens of wrong notes.
- Double, don’t replace. Layering the generated MIDI under the original audio — a sub bass under a vocal, a pad under a guitar line — tends to sound better than swapping one for the other.
- Check the octave first. Pitch detectors occasionally land a whole line an octave off. Select all in the piano roll and transpose before you fix anything else.