Chimera: a simulated human throat that sings and talks in your browser
By Pull This RepoPosted 5 min read
Short answer
Chimera is a free, MIT-licensed voice synthesizer that runs in your browser at chimera.audio. It uses no samples and no AI voice: every sound is air pushed through a simulated airway of lungs, vocal folds, throat, tongue, lips and nose. It speaks typed English, sings on a piano roll, beatboxes and calls like an alligator, seal, elephant or songbird.
Chimera is a free website that works like a fake human throat. Type a sentence and it speaks, write lyrics and it sings, press a pad and it beatboxes, pick a creature and it bellows like an alligator. There are no recordings inside: the sound comes from a simulation of air moving through lungs, vocal folds, throat, mouth, lips and nose. We played every mode on the hosted app, so what follows is how it actually behaved.
What is Chimera?
Chimera is a physical-modeling voice synthesizer by Jameson Hodge, released under MIT. The engine is written in Rust and compiled to WebAssembly, so it runs in a browser tab; the same code runs natively for tests and a command line tool. The repo was created on 5 November 2026. It had 1,881 stars six days later when we made our video, and 2,722 when we checked on 18 November 2026.
The README sums up the approach as "No samples, no neural audio." Everything on screen is drawn from the live simulation: a cutaway of the head, the airflow, the vocal folds slowed down 100 times, the spectrum and the formants.
Try it with no install: open chimera.audio. The old address, chimera-vocal-synth.vercel.app, now redirects there. Browsers block sound until you click once, so press Start and hear a song or Start audio first.
What can Chimera do?
It has five main modes, one per tab:
| Mode | What you do | Try this |
|---|---|---|
| Play | drag the tongue, lips and jaw in the cutaway head, or use the vowel pad | click Hold breath, then drag the green crosshair |
| Speak | type English text or IPA and press Speak | the preset tongue twister "She sells sea shells by the sea shore" |
| Sing | write lyrics and play them on a piano roll | the built-in song Made of air |
| Beatbox | pads on keys 1 to 0 and a step sequencer | the Boots and Cats preset |
| Bestiary | play animal airways | American alligator Bellow |
A few details we found useful:
- Play starts silent because the lungs are at zero. Hold breath or press a piano key to push air through it. Whisper makes the voice breathy.
- Speak shows the phonetic transcription under the box and a gestural score of which part of the mouth moves when. The Rate slider sets the speed.
- Beatbox writes a pattern as text, such as
B . t . | Pf . t ., so you can type a beat instead of clicking boxes. - Bestiary has four creatures with their own airways: a songbird with a two-voice syrinx, a Weddell seal, an American alligator and an African elephant.
Male, Female and Child buttons change the size of the airway, and the README says the bodies blend continuously. A MIDI keyboard works through Web MIDI.
Since our video, Sing mode has gained more. The README says you can type words and a Sing button sings them to a tune picked by their mood, Record video makes a 9:16 clip, and lyrics can be English, Korean or Russian. There is also a Duet view with two 3D heads.
How do you make Chimera say any word?
Respell it the way it sounds, or type it in IPA. Speak mode reads English with the CMU Pronouncing Dictionary and guesses unknown words with letter-to-sound rules, so names and slang often go wrong. Three fixes, easiest first:
- Respell it. "Kai meer uh" instead of "Chimera". Split long words with spaces and add commas for pauses.
- Switch on Text is IPA and type the sounds, for example
həˈloʊfor hello. The consonant chart and vowel buttons on screen show every symbol, so you can copy them. - Slow the Rate slider. Fast consonant clusters smear first.
Check the transcription line before you press Speak. If it shows the wrong sounds, fix the text, not the voice.
How do you run Chimera yourself?
From the README (we did not run these). You need Rust with the wasm32-unknown-unknown target, and Bun:
git clone https://github.com/jamesondh/chimera-vocal-synth
cd chimera-vocal-synth
bun install
bun run devThat builds the WebAssembly and serves the app on 127.0.0.1:5317. The audio engine needs a secure context, so open it on localhost or over HTTPS. The command line tool renders to WAV. The README's line lists the kinds of sound it takes after render; pick one and add its options:
cargo run --release -p chimera-cli -- render speech|text|song|notes|beatbox|creature|vowels …What is the catch?
- It sounds robotic. The README's goal is to get as close to a real voice as the simulation allows, while the robotic sound "stays playable as a preset". It is a physical model, not a voice clone. In our capture, the synth's "Hello" came out rough enough that a transcriber misheard it.
- Speak reads only English text or IPA. Other languages arrive one front end at a time, each needing a native speaker's check.
- It is very new and changes fast. The last push was hours before we recorded, and again on the day we checked, so buttons may move.
- Sound starts only after a click, and a slow laptop can stutter; the top bar shows the engine load.
Who is Chimera for?
It suits musicians who want a strange, hands-on vocal instrument, teachers and students of phonetics who want to see how the mouth shapes sounds, and anyone who wants to make a browser say something odd. If you need a natural-sounding narrator, a neural text to speech tool is still the better pick.
How does Chimera compare with other vocal synths?
| Tool | What it is | Good for |
|---|---|---|
| Chimera | full airway simulation with speech, singing, beatbox and animals | playing a voice as an instrument |
| Pink Trombone (zakaton) | programmable version of the classic mouth toy | simple tract experiments in code |
| voc | a vocal tract physical model written as a readable literate program | learning how such a model works |
| eSpeak NG | classic open source speech synthesizer | text to speech in more than a hundred languages and accents |
What did we run for the video?
We used the repo's own hosted web app in a browser: we played the built-in song, typed our own line ("Hello. I have no lungs.") into Speak mode, dragged the tongue in Play mode, ran a Beatbox preset and played the alligator and songbird in the Bestiary. We recorded the tab's audio. We did not clone, build or run the code ourselves.
For another way to turn sound into code in a browser, see Toaster, which converts music to Strudel. For a free beat maker, see LYNX SHARE, and to let an AI agent check a soundtrack, see Earshot.
Pull it. Stash it.
Questions people ask
Is Chimera AI text to speech?
No. The README says there are no samples and no neural audio. Chimera is a physical model: it simulates air moving through an airway, and the shape of the tongue, lips and throat makes each sound. That is why it can also beatbox and imitate animal airways.
Is Chimera free to use?
Yes. The hosted app at chimera.audio needs no sign-in, and the code is open source under MIT. The CMU Pronouncing Dictionary and fonts bundled with it keep their own licences.
Why does Chimera say some words wrong?
Speak mode looks words up in the CMU Pronouncing Dictionary and guesses unknown ones with letter-to-sound rules, so names and slang often come out wrong. Respell the word the way it sounds, or switch on Text is IPA and type the sounds directly.
Can Chimera sing in other languages?
Since our video, the Sing mode accepts lyrics in English, Korean (Hangul) and Russian, according to the README. Speak mode still reads English text or IPA. Other languages are planned through one shared front end, each checked by a native speaker.
Can I save the sounds I make?
We did not find an export button in the web app when we used it, so we recorded the browser tab's audio. The README now mentions a Record video button in Sing mode that makes a 9:16 clip, and a command line tool that renders speech, songs, beats and creatures to WAV files.
Sources
More repos like this



Get tomorrow's repo in your feed
One free, open source repo a day as a short video, shown working, with the catch named. Pull it. Stash it.
