// VOX EX MACHINA — NO SAMPLES, NO NEURAL NETS
CHIMERA
A voice made of air. Every sound here comes out of a simulated throat: lungs, two vocal folds, a tongue, a jaw, lips and a nose, computed sample by sample in your browser. The throat on the stage is the one you hear.
> PATTERNS — EVERY DRUM IS A MOUTH SHAPE
> OTHER AIRWAYS
> BODY
Loading the airway (about 1 MB)…
How it works
01 / LUNGS Lung pressure pushes air through the glottis, which opens and shuts a hundred-odd times a second. The air coming through it is the source; there is no oscillator playing a waveform.
02 / TRACT The air then crosses a branching tube from the larynx to the lips and nostrils, its walls soft tissue on springs, solved as a wave equation node by node at two to four times the audio rate (an FDTD airway). Its shape is the vowel: the curve under the throat is that tube's resonance, and its peaks are the formants.
03 / GESTURES Words become phonemes (the CMU dictionary, or IPA), phonemes become overlapping gestures, and gestures steer the tongue, jaw, lips and soft palate, each a critically damped spring. Hiss and plosives are the air going turbulent where the tube nearly closes.
CHIMERA is Jameson Hodge's
physical-modelling voice synthesizer (MIT), a Rust engine compiled to
WebAssembly and run in an AudioWorklet; its own home is chimera.audio.
This page is a cartercole.net front end on the same engine, and /chimera/app is
his whole instrument, unchanged, built from cd98192. Pronunciations from the
CMU Pronouncing Dictionary. Nothing you type leaves
the page.