Compare two songs

Why the tool that says «this sounds like…» does not exist

It would be wonderfully handy: you load your track and a program tells you which song in the world it resembles. It does not exist, and not because programmers are lazy. Answering that would need the archive of the world's recordings, which does not fit inside a web page and which nobody is allowed to hand out anyway.

The acoustic fingerprints you know — Shazam's, to be clear — do not solve it either: they recognise the very same recording, not a melody that resembles it. On a cover, on a piece played again or on a song written independently they say nothing. And YouTube's Content ID, which is what people actually fear, is a private archive: nobody queries it from outside.

A tool printing «you are fine» would be the worst thing to put in here: it would reassure someone on a legal question, for free and wrongly.

What this comparison looks at instead

Here there are two tracks and you pick them both: yours and the one you are worried about. The comparison looks at the chromagram, that is how much energy sits in each of the twelve notes at every instant. Timbre, instruments and production disappear, and the musical substance stays.

Two devices go on top, otherwise you measure the arrangement instead of the song. The first: all twelve transpositions are tried, because a piece played a tone higher is the same piece. The second: the alignment is local and copes with a different tempo, so it finds the section that matches instead of demanding that everything match — which is why the report tells you from which minute to which minute.

The melodic contour is measured only when a single voice is playing. In a track with several instruments at once the melody cannot be isolated, and then the page says so rather than printing a number that does not hold. Tempo is always shown but never counted: a slower cover is the same song, and half the record shops sit at 120 beats per minute. If you need to measure it on its own there is the BPM counter.

How to read the number, and what it does not mean

The thresholds are not eyeballed: they are measured on pairs built on purpose, whose truth is known beforehand. The same track transposed, sped up, re-orchestrated and dirtied with drums lands between 92 and 100; two tracks with nothing in common land between 14 and 21; a pair sharing only the chorus falls halfway, around 49. The lines are drawn where the gap is.

What the number does not tell you is whether you have a legal problem. A common chord progression belongs to nobody and pushes the score up even between songs that have nothing to do with each other; a handful of shared notes, on the other hand, has been enough to lose million-dollar cases. Courts decide these things case by case, and the courts themselves do not agree with one another.

Lyrics, timbre and production stay out of it: this comparison looks at the notes, not at the words nor at the sound. And the whole rest of the world stays out too. If you want to check that the file you are about to use is not at the wrong loudness, look at loudness in LUFS.