Riffscribe
Give it a song. Get the sheet music, the tab, and a backing track.
Riffscribe listens to a recording and writes down what it hears — notes on a stave and numbers on a fretboard. Then it strips the vocals out, slows the whole thing down without dropping the pitch, records you playing over the top, and tells you which bars you still need to work on.
Hear it, read it
Basic Pitch — Spotify's note-detection model — listens to the audio and writes out the notes. You get standard notation and tablature side by side, with the frets chosen so your hand barely moves. Guitar, bass, uke, mandolin, banjo, piano, voice, violin, viola, violoncello, double bass, flute, clarinet, sax, trumpet, trombone.
Mute the singer
Two ways. Instant takes about a second and needs no download. The AI split runs Meta's Demucs right here and hands you four separate tracks — vocals, drums, bass, and everything else — so you can silence any of them.
Slow, not slurred
Drop to 25% speed and the key stays exactly where it was. Or move the whole song up or down twelve semitones without touching the tempo. Loop the four bars you keep fluffing.
Play along, keep the take
Record your own part over the backing track. Record it at half speed if you need to — it gets stretched back to full tempo without turning you into a chipmunk — then mix it and export a wav.
In the key you want
Pick the key the part should be written in and the notes and the key signature move together. Accidentals are spelled for that key — E flat in a flat key, not D sharp — and transposing instruments are handled, so a clarinet or trumpet part is written where those players expect to read it.
Fix what it mishears
Automatic transcription is a first draft. Click any note on the staff and move it a semitone or an octave, or take it out — with an undo. No round trip through MuseScore to correct one wrong note.
Tune up first
A tuner with a proper needle, and a drone you can play against, set to the key of whatever you loaded. For intonation the drone is worth more than the meter: a note a few cents out beats audibly against it long before it looks wrong on a dial.
Tells you how it went
Play it, then ask how you did. Your take is marked against the written part: what was clean, what you missed, what sat flat or came in late, whether you rush or drag, and which bars are weakest — with one press to loop those and go again.
Practises the way you do
A count-in so you can come in with the music. A speed trainer that nudges the tempo up 5% every clean pass until you reach full speed. Named sections, so it is “the chorus” and not “somewhere around 2:14”. And the transport stays pinned to the bottom — you are holding an instrument, not a mouse.
Only takes a song apart once
Separating costs minutes, so the stems are kept on your machine and come straight back the next time you open that song — even if you renamed the file. About 12 MB for a four-minute song, not the 340 MB the raw audio would take.
Send it to the other players
Sign in and your chart is saved — parts, tempo, key, sections, loops. Share the link and each player opens the same writing against their own copy of the song. Only the notes travel; the recording never leaves your machine.
Reads along with you
Turn on follow-the-music and the score scrolls itself, highlighting the beat being played, so you can keep both hands on the instrument. The part is played back in your instrument's voice too — mute it once you can hold the line yourself.
A part each, or a whole quartet
Write out as many parts as you like — the bass from one stem, the harmony from another — and they engrave as one score, a staff each. Any part can then be split into voices and handed to a duo, trio, quartet or quintet. That is arranging one line across players, not pulling the original musicians back out of the recording; nothing can do the latter.
Takes your file anywhere
Export MIDI, MusicXML for MuseScore or Guitar Pro, alphaTex, printable PDF, and wav — every part on its own staff, and audio as the whole mix, the backing track, one row of the mixer, or just the bars you looped.
Or just ask
There is an assistant in the corner that works the studio for you. “Split the stems and write the bass part out.” “Slow it to 60% and loop the first eight bars.” It presses the same buttons you would.
Nothing leaves the room
No upload, no queue, no per-minute pricing, and no account needed to transcribe. The models run inside the browser tab, so the song stays on your machine.
Being straight with you about accuracy
Automatic transcription is not a human transcriber. It is very good on a solo instrument, decent on a clean stem, and messy on a dense full mix — so split the stems first and transcribe one at a time. Treat the output as a strong first draft: the rhythm grid, tempo and key are all editable, and the exported MusicXML opens in MuseScore for cleanup.