Chimera Vocal Synth

Chimera Vocal Synth (branded CHIMERA in the app) is a physical-modeling voice synthesizer by Jameson Hodge. Sound comes from simulated lungs, vocal folds or a syrinx, and a branched vocal tract — not from sliced recordings. The engine is written in Rust, compiled to WebAssembly for an AudioWorklet in the browser; the same code also runs natively as the chimera CLI. The repository is https://github.com/jamesondh/chimera-vocal-synth and the demo is https://chimera-vocal-synth.vercel.app/ . GitHub lists the repo as created on 5 November 2026; it had about 1,289 stars at the time of this scan.

According to the repo, you can drag the tongue, lips, and jaw, speak from English text or IPA, sing with lyrics on a piano roll, beatbox from pads and a sequencer, and switch in songbird, seal, alligator, or elephant voices. Male, female, and child bodies blend continuously. The license on the repository is MIT.

This radar database already has ElevenLabs Chimera, which is a different product from a different company. This entry is only Hodge's open-source physical-modeling synthesizer.

Highlights:

• Demo: https://chimera-vocal-synth.vercel.app/ ;
• Source and docs: https://github.com/jamesondh/chimera-vocal-synth ;
• Coverage and discussion on X (and Hacker News, in secondary write-ups) circulated in this window.

Why it matters

Most singing synthesizers stitch recorded phonemes and then shift pitch. CHIMERA takes the other path: it models the airway as a network of tubes so vowels, consonants, and some animal voices come out of geometry and airflow. For teaching, vocal experiments, or an interactive explainer, that is more useful than a clip you can only listen to, because you can drag articulators and watch the tract and formants move.

It also ships as a complete browser app, not just a paper or a CLI. The MIT license, a public repo, and a demo you can open immediately are why the name is worth recording on its own. The name collides with ElevenLabs Chimera, so this page uses Chimera Vocal Synth as the term and names the author and repository.

Limits

The project itself says naturalness is still being tuned, and consonants, timing, and real vocal-fold motion are harder than held vowels. A secondary article recaps a project-side comparison in which Whisper word error stays much higher than on real speakers; that is not our own test. Star counts change; use the GitHub page at the time you look. This is an independent open-source project, not an ElevenLabs product, and you should not assume commercial dubbing stability or broad language coverage.

1. Open https://chimera-vocal-synth.vercel.app/ and try dragging the tract, speaking, or singing (it needs a secure context: localhost or HTTPS);
2. For a local build or the CLI, follow https://github.com/jamesondh/chimera-vocal-synth : you need Rust with the wasm32 target, plus Bun, then the repo's dev or test commands;
3. Evaluate it as a physical-modeling and teaching tool, not as a default commercial TTS or singing engine;
4. Keep the MIT license in mind if you reuse the code, and do not describe it as the same product as ElevenLabs Chimera.