The Music of the Families
Tuning 1,166 documented sound systems…
Babel · The Music of the Families

Phonology is inherited. You can hear it.

A language's sound system isn't random taste — it descends, like its words. Here every one of 1,166 documented inventories becomes a playable sonic fingerprint, placed by family so siblings cluster. Open a family and play the whole clade: the shared signature plays as one drone, each member a variation. Then find the outliers — the languages that sound nothing like their kin, because contact overrode descent.

An artistic sonification of documented phoneme inventories — not phonetics. It will not teach you to pronounce anything. Sounds come from PHOIBLE: one documented inventory per language, so a clade's "mean" and its outliers are sample-bounded, and "rarity" is rarity-in PHOIBLE (skewed toward well-studied regions), not rarity on Earth.
The documented world
Tap a language on the globe or open a family to hear it sing.
silent
headphones recommended · audio starts on your first tap
The method — how a phoneme inventory becomes music, and what it can’t tell you

For each of 1,166 languages we take its documented PHOIBLE inventory — the actual phoneme set a linguist catalogued — and read five things off it, each mapped to one musical axis:

  • Vowel share → openness of the ambient pad. Polynesian openness breathes; Northwest-Caucasian density is taut.
  • Mean world-rarity of its segments → timbre. Common /m i k/ = warm; rare ejectives, uvulars and clicks = metallic, glassy.
  • Inventory size → density of events. !Xóõ (161 segments) teems; an 11-segment language is sparse.
  • Tone (any tone segment) → a pitched glide layer.
  • Clicks → a literal click-percussion layer.

The signature act — "Play the family tree": a clade's mean inventory becomes a sustained base drone, and each member layers in as a variation panned by its real geography. Members that hug the mean blend in; outliers stab out. That is genealogy you can hear.

What this is honest about: PHOIBLE samples one documented doculect per language, so a "family mean" and an "outlier" are sample-bounded, not ground truth. Rarity is rarity-in-PHOIBLE — documentation is uneven, so the rarity map can under- or over-light regions by study bias, not true rarity. 1,113 of 1,166 languages have a Glottolog classification; the other 53 sit in an honest Unclassified branch, never invented into a family. Word-melodies are real wikipron IPA for the 116 languages that have it — everyone else shows "no transcribed words on file", never a faked pronunciation. And 7 famous tongues from the old plate (Hawaiian, Turkish, Polish, French, Navajo, Georgian, Chinese) have no PHOIBLE inventory at all — they cannot be sourced this way, and we say so rather than substitute.

Data: data/music_reborn.json (fingerprints, tree, clade centroids, outlier ranks, auto-blurbs) and data/music_audio/<code>.json (real wikipron words, lazy-loaded), built from PHOIBLE + Glottolog + wikipron. Synthesis is pure Web Audio — no audio files shipped.