Audio to sheet music: a melody, written out.

At a glance

How a recording becomes a score

The page follows the pitch of the recording a hundred times a second with the YIN method, the same tracker behind the site's tuner, which reads within a cent from a low B to a high E. Where the pitch holds steady for long enough it marks a note, and where a held pitch is played again (two quarter notes on the same G, say) it splits the note at the dip and re-attack in the loudness, because a pitch tracker alone hears one long note there.

The tempo comes from the notes themselves. Every possible beat length is tried, and the one that lines the note starts up best wins; that also gives where beat one falls. Each note is then placed on a grid of sixteenths, written with the plain and dotted values a reader expects, and tied across bar lines and beats where it has to be, so every bar adds up. The key signature comes from the site's key finder, which weighs every note against the 24 major and minor keys, and decides whether accidentals are written as sharps or flats.

Getting a clean result

Record one line, as dry as you can: a phone close to the singer or instrument, in a quiet room, without a backing track. Play or sing in time; the tempo is found from your notes, so a steady pulse is what makes the rhythms come out as you meant them. If the result is written at double or half speed (eighth notes where you meant quarters), type the tempo you had in mind and the whole score redraws. Choose the time signature the same way; the page cannot hear where a bar starts, so it begins bar one on the first beat.

For a song with a band, isolate the voice first with the stem splitter and bring the vocal stem here. For a melody you would rather edit note by note, use audio to MIDI, which has a piano roll, then open the MIDI in MuseScore.

For your instrument

"Written for" transposes the part so it reads correctly on a transposing instrument: a B♭ trumpet or clarinet part sits a tone above concert pitch, an alto sax part a major sixth above, a horn part a fifth above, and the key signature moves with it. Guitar music is written an octave above where it sounds, and the octave option does that. The clef follows the range of the melody unless you choose one.

What it can't do

This writes out one line. Chords, piano with both hands, and a full mix have no single note to follow, so they come out as a scatter of the loudest notes or nothing at all; for the chords of a song, use the chord finder. Ornaments, slides and vibrato are written as the nearest plain note. Triplets and swing are put on the straight sixteenth grid, so a swung tune reads as dotted rhythms. The figures above were measured on synthesized melodies, where the notes are known exactly; a real singer will need more corrections, and the MusicXML is there so you can make them in a notation program rather than starting from a blank page.