CHIMERA makes browser music by letting you rearrange a throat

A virtual head has its mouth open, and you’re allowed to rearrange the tongue while it makes noise. CHIMERA gives the expression “watch your mouth” an unusually literal interface.

The maker’s README describes a playable cutaway head, with keyboard and MIDI control alongside the direct dragging. Moving anatomy changes the simulated instrument that produces the voice. That puts the interesting part where a curious person can poke it, with considerably less paperwork than a real dissection.

The throat has plumbing. CHIMERA’s acoustic design describes a branching network in which air pressure and flow interact with openings and constrictions. A closed passage stops flow; a narrowing changes the conditions for noise. The shapes in the head have jobs beyond looking appropriately medical.

There are limits to the anatomical ambition. The design document explicitly separates unfinished ideas, including a more elaborate vocal-fold model, from what exists. Its own warning is “Treat the code as the truth” when the design disagrees. An ambitious diagram can get ahead of its implementation (shocking, I know), and this one says where to be careful.

Singing brings a lovely timing detail. The README says “consonants land before the beat.” A piano roll can pair lyrics with notes, but preparing a syllable takes movement. Even a software singer has to enunciate, and preparing the consonant before the note treats the mouth as something that needs time to move.

The browser runs the Rust engine through WebAssembly in an AudioWorklet. The project’s engine documentation also describes native use and command-line rendering. Keeping those routes together means a sound explored through the head can belong to an engine with a life beyond the browser tab.

Real-time audio has unglamorous rules. The engine document separates rendering from operations that allocate memory, and records a blunt failure behavior: a panic in the audio worklet produces silence until a new node starts. An instrument has to keep delivering samples while the attractive diagrams do their thing. The reset behavior is worth knowing before treating a browser experiment as a dependable performance rig.

A physical model invites a different sort of exploration. Hold a note, change the mouth, and the resulting difference has a visible cause. That could make CHIMERA useful for demonstrations as well as peculiar music, provided its simulated anatomy is presented as a model with documented limits. A cutaway head doesn’t become a clinical authority just because its mouth moves.

The project’s documentation is the source for these capabilities; this is not a listening test or a claim that the synthesized voice passes for a human singer. Further work could make the same exposed controls easier to use musically without hiding the mechanism. For now, the documentation still leaves in-app audio export unfinished, even as the head gets to sing.

Anatomy class got loud.