Voice to MIDI — hum it, beatbox it, hear it on real instruments
That melody stuck in your head, the beat you just mouthed — record it, tag what each sound was meant to be, and play your exact performance back on real drums and cleared instruments. Built for producers and composers who think faster than they can program. Free in your browser, and your voice never leaves your device.
Hum a melody, beatbox a groove, or both in one take (up to 60s). Recording to the metronome lines the grid up in bars; without it we estimate the tempo from your hits. Your audio never leaves this browser.
How it works
- Record your idea. Hum a melody, beatbox a groove, or both in one take — up to 60 seconds, straight into the mic. An optional metronome keeps the grid in bars. Nothing is uploaded.
- It maps itself. Every note and hit appears as a block on an editable grid — already grouped by how each sound actually sounds, with a real drum matched to each group and an instrument on your melody. Press play immediately.
- Correct anything. Click any block to hear that exact slice of your voice back. Re-tag blocks, rename lanes, or swap any sound — the auto-mapping is a suggestion, and you always have the final say.
- Shape each part. Every lane is a channel strip: level, reverb, and echo per lane, plus velocity, length, timing, and quantize per note. Drag a block longer and the sound genuinely plays longer.
- Export the folder. Download the rendered WAV, the MIDI, and a certificate of creation crediting every cleared pack you used — your performance, your melody, real instruments.
How the auto-mapping stays honest
The tool never pretends to know what a mouth sound “is” — it groups your hits by how they actually sound, so every boom lands on one lane and every tss on another, then picks the real drum whose character is closest to yours. Names like “Kick” and “Hi-hat” are suggestions; the matched sound is what makes the first play feel right. And because clicking any block replays that slice of your recording, correcting a guess takes seconds — you always have the final say.
Frequently asked questions
How does voice to MIDI work?
Two detectors read your recording at once: a pitch tracker catches hummed and sung notes, and a transient detector catches percussive sounds like beatboxed kicks and hats. Your hits are then grouped by acoustic similarity — every "boom" you made lands on one lane, every "tss" on another — and each lane is matched to the closest-sounding real drum, so the mapped mix plays instantly. Everything is editable if it guessed a name wrong.
Is it free? Does my voice get uploaded?
The whole tool runs in your browser — recording, analysis, labeling, mapping, and playback are free, and your voice never leaves your device. Exporting the finished WAV + MIDI + certificate uses a download like the rest of Selekt.
What sounds can my voice play through?
Drum labels map to real one-shots (808, acoustic, and lo-fi kits) or any instrument pack triggered per hit; melody labels play through cleared CC0/public-domain instrument packs — pianos, strings, brass, and more — keeping your exact pitches and timing.
Who owns what I make with it?
The melody and rhythm are your own performance — you made them. The sounds are cleared Selekt packs and kits, and the export includes a certificate of creation crediting every pack used, so you have the provenance in writing.
Do I need to sing well for this to work?
No. Close-mic, deliberate sounds convert best — but the grid is fully editable, so you can fix any wobbly pitch, stretch a note you meant to hold longer, nudge timing onto the grid, and delete anything stray before you map sounds.
Have a stem or a finished recording instead of a voice memo? The Audio to MIDI converter reads instruments, and the stem separator splits whole songs into parts, MIDI, and chords.
