The companion to things that do not ask to be heard, and the more interesting of the pair, because it refuses to be an improvement on it. Same stack — NumPy, SciPy, SoundFile, Pedalboard — and a deliberately different theory of what a sound is.
The first work is made of events that begin: something is struck, it rings, it decays. Every material in it, however different in timbre, belongs to that one family. Codex's response was not to add better timbres but to cancel the premise. If sounds do not begin, then a mouth is always already open and the composition can only be about what is permitted through — which turns rhythm from a matter of striking into a matter of shutters, permissions and erasure.
Together the two establish the finding that the equation-first ASCII films then build on: the ceiling on generated sound is not the package list or the audio quality. It is the model's repertoire of source worlds and temporal laws. Changing timbre alone does not change the ontology; you have to change what exists, how time operates, or what the listener is allowed to hear.
No words are spoken — the material is vocal in timbre but never articulates language. Forty-one seconds of five continuous synthetic mouths, each drifting independently between the formant shapes of a, e, i, o and u, so the sound behaves like vowels sliding into one another without ever landing on a syllable. Nothing starts: the streams are already running when the file opens. What produces rhythm is interruption — seventeen-step shutters running at five rates that never line up, chopping the voices into shifting polyrhythms, and section changes that revoke or grant permission for particular mouths to be audible at all. A separate band of unvoiced air supplies sibilant, breathy articulation between them. Eight abrupt full-spectrum erasures cut the whole field to nothing, more like splices in tape than rests. The streams move slowly through the stereo field. The piece ends by being cut off rather than by finishing.