A physical-modeling vocal synthesizer in Rust/WebAssembly that sings, speaks, beatboxes and imitates animals from a simulated airway.
- Language
- Rust
- License
- MIT
- Stars
- 1.2K
- Try it
rustup target add wasm32-unknown-unknown && git clone https://github.com/jamesondh/chimera-vocal-synth && cd chimera-vocal-synth && bun install && bun run dev
Chimera is an open-source vocal synthesizer that makes speech and song without a single recorded sample. Instead it simulates the body that produces a voice: lungs, vocal folds (or a bird's syrinx), a branched vocal tract, lips and a nose. You can grab the tongue, lips and jaw on a cutaway head and drag them while it sounds. Developer Jameson Hodge published the repository on November 5, and by November 10 Trendshift counted about 1,240 new stars and 102 forks, with the biggest single day on November 9 (+592). A Hacker News post linking the live demo reached 339 points.
Why it matters: most voice tools today are either sample libraries or neural text-to-speech models trained on recordings. A physical model is a different trade. It sounds less polished, but every parameter is something you can see and touch, which makes it a playground for musicians, phonetics students and anyone curious how a voice works. It also runs entirely in the browser.
What it does
The README and The Audio File's write-up describe six modes:
| Mode | What you get |
|---|---|
| Play | Drag tongue, lips and jaw live; vowel pad, encoders, computer keyboard and Web MIDI |
| Speak | English text (CMUdict dictionary with a letter-to-sound fallback) or IPA, turned into articulatory gestures and prosody |
| Sing | A piano roll with lyrics per note; consonants are placed just before the beat so the vowel lands on time |
| Beatbox | Pads and a step sequencer, with each drum sound modelled as a mouth gesture |
| Bestiary | Songbird (two-voice syrinx), seal, alligator and elephant |
| Duet | Two 3D heads replay a pre-rendered song |
Male, female and child "voice bodies" blend continuously rather than switching presets, and the on-screen visuals are driven by live engine state, including airway pressure, airflow, vocal-fold motion, spectra and formants.
How it's built
The engine is written in Rust and compiled to WebAssembly that runs inside a browser AudioWorklet. The same code also builds natively, which gives the project unit tests and a chimera command-line tool for offline rendering and diagnostics. The repo's docs refer to a finite-difference time-domain (FDTD) airway model, a real-time and determinism contract, and gesture-based articulation. They don't publish latency or CPU figures, so there's nothing to benchmark against yet.
The repository includes AGENTS.md and CLAUDE.md files and a docs/hill-climb/ folder holding rules and a log for tuning the voice in measured steps, which suggests AI coding agents were part of the workflow. The README doesn't spell out how they were used. GitHub's API lists the license as MIT.
Try it
The fastest route is the hosted demo linked from the repo (it needs a secure context, so HTTPS or localhost). To run it locally, the README asks for Rust with the wasm32-unknown-unknown target and the Bun runtime:
rustup target add wasm32-unknown-unknown
git clone https://github.com/jamesondh/chimera-vocal-synth
cd chimera-vocal-synth
bun install
bun run dev # Vite dev server on 127.0.0.1:5317
bun run test
The CLI renders to audio files with cargo run --release -p chimera-cli -- render ....
What commenters liked, and what's rough
On Hacker News, users posted overtone-singing experiments and song covers and called the setup impressive. Many pointed to Pink Trombone (2017), the much-loved browser vocal-tract toy, and to Bell Labs' 1939 Voder as ancestors. The complaints were practical:
- Extreme tongue, lip and soft-palate positions can produce a screech.
- Several people hit iOS scrolling problems and noisy or slow loading on phones; MadeByWho says mobile compatibility hasn't been confirmed.
- Some browsers lack Web MIDI, which limits controller use.
- Music producers asked for a VST or CLAP plugin; today it's browser and CLI only.
- Duet mode only replays pre-rendered songs.
Who should care
Musicians and sound designers looking for strange, controllable vocal textures; teachers who want to show how a vowel is formed; and developers interested in real-time DSP in Rust and WebAssembly. Don't expect natural text-to-speech; expect an instrument. The project is five days old and moving fast (Trendshift logged 19 merged pull requests in that span), so pin a commit if you build on it.
Sources
- GitHub — jamesondh/chimera-vocal-synth github.com
- Hacker News — Chimera Vocal Synth news.ycombinator.com
- The Audio File — CHIMERA ~ Physical-Modeling Vocal Synthesizer audio-file.org
- Trendshift — jamesondh/chimera-vocal-synth trending stats trendshift.io
- MadeByWho — CHIMERA vocal synthesizer by Jameson Hodge madebywho.com