Mouth Synth

Speech is not made of samples, it is made of a tube. Air from the lungs sets the vocal folds buzzing at the glottis, and the shape of the tract above them — the tongue, the lips, the soft palate — filters that buzz into the vowels and consonants we recognise. Mouth Synth models exactly that: a glottal source driving a one-dimensional acoustic waveguide, the same physical approach Neil Thapen's Pink Trombone made famous, rebuilt here as a playable instrument rather than a toy.

The tract is a chain of short tube sections, each with its own cross-sectional area, connected as a Kelly-Lochbaum digital waveguide: sound reflects at every area change and travels back and forth, and the standing waves that survive are the formants you hear as a vowel. Moving the Tongue controls slides the tightest point of that tube forward and back and raises or lowers it, which is the whole difference between 'aah', 'eee' and 'ooh'. A separate Constriction pinches the tube somewhere specific — near the teeth for an 's', at the lips for an 'f' — and when it closes fully and reopens the model fires the little pressure transient that makes a 'p' or a 't' pop.

Because it is one tube it is monophonic, and that is a feature: like a real voice it cannot sing a chord, so moving between notes slides the pitch through everything in between. The Glide control sets how fast, from a snappy step to a long portamento wail. Tenseness and Breath trade the source between a pressed, buzzy fold vibration and an airy whisper, and the nasal branch — opened by the Nasal control's velum — lets sound leak through the nose for the hums and the 'm'/'n'/'ng' family.

This is the browser preview lane. The audible DSP is the restricted-TypeScript voice-mouth engine run in an AudioWorklet — the same source used by optimized browser and plugin builds — so nothing here is a throwaway. Its reverb is a bounded Freeverb inlined into the voice so the tail ships with it. And because the tract is just a set of numbers you can drive from anywhere, the module's camera mode maps your own mouth's openness, straight from MediaPipe face landmarks, onto the model's controls — the same landmark plumbing the Hand Mod module uses.