Glossolalia BEN BURRON
Human
Alien

How it works

There are no recordings in here. Every voice is a buzz, like vocal folds, pushed through a model of a throat whose shape is rewritten every five milliseconds, then calculated one sample at a time.

Before a word is spoken, each voice runs it through its own made-up language. The same word always comes out the same way, which is why the babble sounds like it means something. Every figure below is drawn live from the engine, for the voice you have picked.

01From letters to babble
02How much English survives
The share of each sentence’s sounds that come out unchanged, measured from the engine across the dial. Pip never leaves English; the creatures lose most of it by halfway. It measures sounds, not understanding.
03The throat as a filter
Five resonators in a row, set to the engine’s own values for three vowels. The same buzz goes in; where the peaks sit is the whole difference between “ee”, “ah” and “oo”.
04How many languages it can speak
—different languages from the language switches alone, before a single dial is turned
05The signal path
Every clip takes this route, one sample at a time. The dial moves the presence lift: towards Human it pushes up to 4 dB around 2.2 kHz, where the ear finds consonants, and trims the low end to match.
44.1 kHz, one sample at a timea new throat every 5 ms5 resonances in cascade16 voices53 settingsrendered in your browser

It began as a recreation of the babble voices in Animal Crossing. The first version wired Web Audio filters together and could never sound like a person, so the engine was rewritten as a sample-by-sample Klatt synthesizer, the approach behind the speech machines of the 1980s. Human speech turned out to be just one point on its range.