Hold your phone up to a speaker and within seconds Shazam tells you the song and the artist. It felt like science fiction when it arrived, and it quietly became something we all expect. But notice what it answers. It tells you what the song is. For a listener, that is the whole question. For a musician, it is barely the start, because the question that actually matters is the next one: how do I play this? That question is harder, and it is finally being answered.
What Shazam actually does, and does not do
Shazam works by fingerprinting. It reduces a few seconds of audio to a compact pattern of distinctive points, then matches that pattern against a huge database of fingerprints from known recordings. When it finds the match, it hands you the metadata: title, artist, album. What it never does is listen to the music in any musical sense. It does not know the melody, the chords, the key, or a single note. It is a brilliant lookup, not an act of hearing. That is why Shazam can name a song it has on file but can tell you nothing about a tune you hummed yourself, and nothing about how the notes go.
The harder question: how do I play it?
Telling you how to play a song is a different problem entirely, called transcription, and it is genuinely harder. There is no database to look the answer up in. The software has to listen to the recording and pull the music out of it: separate the melody from the accompaniment, work out which notes are sounding when several overlap, decide the rhythm, infer the key, and lay it all out as notation. That is analysis, not matching. It is closer to what a trained musician does with headphones and a pause button than to what a fingerprint does.
For a long time it was too hard to do well, which is why generations of players learned songs by ear the slow way, or paid for sheet music if it existed at all. What changed is that purpose-built transcription models can now do a strong first pass automatically, and this is what Songscription was built for. Give it a recording, in the browser, with nothing to install, and it isolates the part you choose, then returns the notes as a score and a piano roll, with the chords detected along the way. The answer to how do I play it is no longer locked inside the recording or inside an expert’s ear.
A Shazam for sheet music, with one honest difference
The shorthand we use for Songscription is a Shazam for sheet music, and it is a useful picture as long as you keep one difference in mind. Shazam’s answer is either right or it is not; a fingerprint either matches or it does not. Transcription’s answer is a reading of the music, and like any reading it can be more or less accurate depending on what it is reading. Songscription’s models were trained from the ground up on real musical performances rather than adapted from speech, which is why a clean, sparse recording transcribes beautifully. A dense, heavily produced mix is harder, and the result will have spots to fix. So the honest version of the promise is not a flawless score in one tap. It is a strong, editable draft of how the song goes, produced in a minute or two, that you then correct against your ear.
That distinction is not a disappointment, it is the point. A draft you can read, slow down, and play is an enormous head start over a recording you can only listen to. You go from staring at a locked audio file to sitting at the instrument working through the actual notes, which is the thing you wanted all along. If you want the realistic picture of where accuracy holds up and where it slips, we wrote about what to expect from transcription accuracy.
From identifying music to understanding it
There is a nice arc here. The first wave of music technology helped us find and identify music: search, recommendation, Shazam. The interesting frontier now is technology that helps us understand and make it, that turns a passive recording into something you can read, learn, change, and play. Identifying a song answers a moment of curiosity. Learning to play it is a relationship with the music that can last for years. The good news for anyone who ever held their phone to a speaker and wished it could do more is that the more is here. You can take almost any song you love and turn it into sheet music you can actually play.
Frequently Asked Questions
Is there a Shazam for sheet music?
Sort of, but it works differently from Shazam. Shazam identifies a song by matching a short audio fingerprint against a database; it tells you the title and artist but nothing about the notes. Getting the sheet music is a different and harder problem called transcription, where software listens to the recording and works out the actual pitches and rhythms, then writes them as notation. Tools like Songscription do this: you give them an audio file or a link and they produce a score, a piano roll, and chords you can read and play.
Why is telling you how to play a song harder than identifying it?
Identifying a song only requires matching a unique acoustic fingerprint to a known recording, which does not involve understanding the music at all. Telling you how to play it requires pulling individual notes out of a mix where instruments overlap, separating melody from harmony, and deciding the rhythm and key. That is a genuine analysis of the sound rather than a lookup, which is why transcription results are a strong draft to correct rather than a guaranteed-perfect answer.
Can AI tell you the chords and notes of any song?
It can produce them for most recordings, with accuracy depending on the source. A clean, sparse recording transcribes well; a dense, heavily produced mix is harder and will need more correction. The realistic expectation is a first draft of the notes and chords that you review against the recording and fix, rather than a flawless score on the first try. For learning a song, even an imperfect draft you can slow down and play is a huge head start.
Next time a song gets stuck in your head, do more than name it. Find out how to play it.