Jameson Hodge's CHIMERA Simulates a Full Vocal Tract Inside Your Browser
CHIMERA is a browser-based voice synthesizer that builds sound from a simulated airway, not samples or neural nets. Rust engine compiled to WebAssembly runs in an AudioWorklet, with the same code usable natively via a CLI.
Takeaways
- CHIMERA is a browser-based voice synthesizer that builds sound from a simulated airway, not samples or neural nets.
- Rust engine compiled to WebAssembly runs in an AudioWorklet, with the same code usable natively via a CLI.
- Supports speech from text or IPA, singing with a piano roll, beatboxing, and animal voices like songbirds and seals.
- Male, female, and child bodies blend continuously; live cutaway UI animates from actual engine pressure and airflow state.
- Uses FDTD acoustic simulation of the vocal tract, tuned via a logged hill-climbing process against real recordings.
- Open source MIT, free to try at chimera-vocal-synth.vercel.app .
Developer Jameson Hodge has released CHIMERA , an MIT-licensed voice synthesizer that models lungs, vocal folds, airflow and a shapeable vocal tract in real time. The engine runs in a browser through WebAssembly and an AudioWorklet, generating sound without recorded voice samples or a neural vocoder.
A social-media demo featuring a cover of The Lion Sleeps Tonight has highlighted the range of the model, including songbird, seal and thunder-like timbres. The free browser demo runs in a secure context, while the same Rust engine supports native tests, command-line rendering and acoustic measurements.
The Browser Becomes a Vocal Tract
CHIMERA generates sound from a physical model of an airway. Simulated lungs drive air through vocal folds or a syrinx, then through a branched vocal tract with adjustable tongue, jaw, lips and nasal coupling. Changes to that geometry alter the resonant frequencies, known as formants, that shape vowels and other vocal sounds.
The synthesis core is written in Rust and compiled to WebAssembly for the browser. An AudioWorklet runs the engine within the browser’s audio-rendering system, where it must produce each buffer on time to avoid clicks or gaps. The same core compiles natively for automated tests and the project’s CLI.
Five Ways to Play It
The web interface exposes the physical model through five performance modes:
- Play: Drag the tongue, lips and jaw in a cutaway head, or control the voice through a vowel pad, rotary controls, the computer keyboard and Web MIDI.
Reporting is indexed from AlphaSignal. Rights remain with the original publisher and cited sources.
Read original report
