How Signal Match works

The short version: real audio measurements, compared mathematically. No LLM, no listening history, not Spotify's own recommendation algorithm.

The idea

Pick an artist you already like and a second artist you're curious about. Signal Match pulls real audio-feature data for both -- one number per song per dimension, things like tempo and energy -- and ranks the second artist's whole catalog by how closely each song's numbers match the profile of the first artist's songs. The output is a plain ranked list, not a chart position or a popularity score.

The eleven dimensions

Every song is measured on the same eleven audio dimensions, in this fixed order -- it's the order the fingerprint ticks under each result are drawn in, left to right:

  1. Tempo -- speed, in beats per minute.
  2. Energy -- how intense and fast-feeling the track is.
  3. Loudness -- overall volume, in decibels.
  4. Danceability -- how suited the rhythm is to dancing.
  5. Acousticness -- how likely the track is acoustic rather than electric/produced.
  6. Instrumentalness -- how likely the track has no vocals.
  7. Liveness -- how likely it's a live recording versus a studio one.
  8. Speechiness -- how much of the track is spoken word versus sung/played.
  9. Valence -- how positive or upbeat the track sounds.
  10. Key -- the musical key the track is in.
  11. Mode -- major or minor tonality.

The math

Each dimension gets standardized (z-score normalized) against the pool of songs being compared, so a dimension with naturally huge swings -- tempo can range from 60 to 200, loudness is negative decibels -- can't silently dominate a dimension that only ranges 0 to 1. Once everything's on the same scale, each candidate song's normalized vector is compared to your taste artist's profile using cosine similarity, a standard way of measuring how closely two sets of numbers point in the same direction. That similarity score is what the match bars and the "Top N%" figure next to each result are built from.

"Top N%" is a percentile within this specific comparison -- rank divided by how many songs were actually analyzed for that pairing -- not a claim about where a song stands against every song ever measured. There's no fixed global reference distribution to compare against, so that's the honest, defensible version of the number.

The fingerprint ticks

Under each result, eleven small ticks -- one per dimension above, always in that order -- show which specific dimensions drove the match. A filled tick means that dimension was close to your taste artist's profile; an outlined tick means it wasn't a big factor. Hover or tap a tick to see the real numbers behind it.

What this can't do

This only measures production characteristics -- tempo, loudness, and the rest above. It knows nothing about lyrics, genre labels, cultural context, or what a song is actually "about," so a mathematically close match can still feel like a strange recommendation, and a genre-obvious match can score lower than expected if the production style differs. That's a real limitation of measuring sound rather than meaning, not a bug to hide.

Data source

Track and audio-feature data comes from the ReccoBeats API, which in turn aggregates catalog metadata from Spotify. See the Privacy Policy for more on what this site does and doesn't collect.

← Back to search