Your characters want to talk. Your budget doesn't include voice actors. Babble Engine turns any line of text into gibberish speech: critter chatter, made-up languages, grumbling trolls, beeping robots, whispering fairies and soft musical tones.
🚫 No voice actors 🚫 No server 🚫 No dependencies ✅ Free, also for commercial games
🎙️ The tool
babble.wrongfloor.studio runs in your browser. Type a line, pick a style and a voice, set the mood, and download a WAV.
- 🐹 16 styles, 70 voices, from Critter Chatter to retro text blips and a talking kazoo
- 😄 6 moods: neutral, happy, angry, sad, flirty, surprised
- 🎛️ Tune every voice: pitch, tempo, expression, throat size, breath, brightness
- 🎤 Record your own voice and your characters babble with it: a sound no other game has. The basic round is under 2 minutes of reading and works with all 12 voice styles. Optional short rounds add double consonants (bl, kr), word endings (ahp, ohm) and barks, and you can redo any round. Recordings stay in your browser, nothing is uploaded. Export a voice to use it in your own game
- 🧑🤝🧑 Save characters so Grandpa Hank sounds like Grandpa Hank every time
- 📜 Script your dialogue: scenes, characters and lines in one place. Record your lines hands-free with a prompter that follows your voice, let babble characters answer in between, speak one word yourself inside a babble line, and give each character its own sound (radio, robot, fairy, giant). Record as many takes as you like and pick the one that goes in. Export one zip for your game (a WAV per line plus
script.json, see lines from a script), or a single line. Import a script from your writer or AI assistant as a file or plain text, and back up your takes
- 💾 Download WAV, ready for any engine
🧩 Web module
Making a web game? Use the same engine straight from your code. One file, no build step.
<script type="module">
import { render, play } from 'https://cdn.jsdelivr.net/gh/wrongfloor-dev/babble-engine@1/site/src/babble.mjs';
const line = render('Hi! Do you want an apple?', { style: 'animal', voice: 'fern', mood: 'happy' });
document.querySelector('button').onclick = () => play(line);
</script>
@1 follows the newest 1.x release. Pin an exact version (@1.0.0) for a game you ship.
- 📜 Plain script tag: load
site/src/engine.js and use the global Babble (Babble.render(...))
- 📦 Node or a bundler:
npm i github:wrongfloor-dev/babble-engine, then import { render, toWav } from 'babble-engine'
- 🔷 TypeScript: types ship in
site/src/engine.d.ts
- ‹/› From the tool: click ‹/› next to Download WAV to get the
render call for the voice and settings you hear
- 🤖 With an AI assistant: babble.wrongfloor.studio/game has a ready prompt for ChatGPT, Claude, Gemini, Cursor or Copilot, and llms.txt lists every option, style and voice id
render(text, options)
Returns { pcm, sampleRate, duration, syllables, reveal, style, voice }. pcm is a mono Float32Array, dry (no room).
| Option | Default | |
|---|
style | 'animal' | see the styles below |
voice | the style's first voice | |
mood | 'neutral' | neutral, happy, angry, sad, flirty, surprised |
seed | 1 | 🎲 same text, options and seed give the same sound. Change it for a different take or another made-up language |
sampleRate | 48000 | |
pitch, speed, expression, throat, breath, brightness | from the voice | override the voice |
hiss | 1 | consonant noise, 0 to 1 |
ownVoice, steady | none, 1 | a recorded voice from loadVoice(), see below. steady 0 follows your recorded pitch, 1 holds each syllable on the melody |
name | the voice's name | Name Talk only: the name the creature keeps saying, like 'Pikachu' |
intent, length | 'tell', 'short' | Bloopa Doopa only: it speaks by intent (greet, ask, joy, letdown…) and ignores the text |
👄 syllables is a list of { time, text }: the start of every syllable in seconds. Flap a mouth or bounce a sprite in sync.
⌨️ reveal has one time per character of your text: when it should show in a text box that types along. Each word types out while it is spoken and waits through the pauses, so the text never runs ahead of the voice. play does the timing for you:
play(line, { onReveal: n => box.textContent = text.slice(0, n) });
🎤 Your own voice in your game
- Record your voice in the tool, under Voices & characters
- Click Export next to it. You get a
.babble-voice.json file
- Load it and babble with it in any voice style:
import { render, play, loadVoice } from 'https://cdn.jsdelivr.net/gh/wrongfloor-dev/babble-engine@1/site/src/babble.mjs';
const me = loadVoice(await (await fetch('me.babble-voice.json')).text());
play(render('Welcome to my game!', { style: 'sim', ownVoice: me }));
The file holds every recorded sound as 16-bit audio in JSON, including the extra rounds, so other engines can read it too.
More
- ▶️
play(result, { room = true, context, onReveal }): browser only. Plays the line with the style's room, like the tool. onReveal(n) is called whenever the first n characters should show. Returns { stop(), ended, context }
- 💾
toWav(result): a 16-bit mono WAV as an ArrayBuffer, dry
- 📋
styles(): every style with its voices, moods and intents
- 🎤
loadVoice(file): a .babble-voice.json from the tool, as text or parsed JSON
📜 Lines from a script
The Script tab in the tool exports a zip for your game: one <line id>.wav per line and script.json (format in llms.txt). Play a line by its id:
const script = await (await fetch('voice/script.json')).json();
const line = id => new Audio('voice/' + script.lines.find(l => l.id === id).file).play();
In Godot 4, put the folder in res://voice/:
var script_data = JSON.parse_string(FileAccess.get_file_as_string("res://voice/script.json"))
for line in script_data.lines:
if line.file:
$Player.stream = load("res://voice/" + line.file)
🎮 Godot 4
Download babble-engine-godot.zip and unzip it in your project folder, or install it from the Godot Asset Library (AssetLib tab in the editor, search "Babble Engine"), or copy addons/babble_engine from this repo. There is no plugin to enable: you get the BabbleSpeaker node and BabbleEngine.
# BabbleSpeaker is an AudioStreamPlayer: pick style, voice and mood in the Inspector
$Speaker.speak("Hi! Do you want an apple?")
$Speaker.syllable.connect(func(index, text): $Mouth.flap())
- ⌨️ Typing text box: set Type Label to a
Label or RichTextLabel and the line types itself into it, in step with the voice
- 📣 Signals:
line_started(text), syllable(index, text), revealed(count) and line_finished
- 🧵 No stutter: the speaker renders on a background thread and starts the line when it's ready
- 🎤 Your own voice: set Own Voice File to the
.babble-voice.json from Export
- 🏛️ Room: add a Reverb effect to the speaker's audio bus
- 🧪 Demo: open
addons/babble_engine/demo/demo.tscn and press F6: pick a style, voice and mood, and watch a mouth move on every syllable
- 🔧 Without the node:
BabbleEngine.render(text, options) takes the same options in snake_case (sample_rate, own_voice) and returns { pcm, sample_rate, duration, syllables, reveal }. BabbleEngine.load_voice(text) reads a voice file. A bad option prints an error and returns {}
Same text, options and seed give the same sound as the web module, sample for sample (tests/godot.mjs checks all 16 styles on every push). GDScript renders about 15 times slower than the browser, which is why the speaker uses a thread.
🐾 Styles and voices
| Style | Name | Voices |
|---|
| 🐹 | animal | Critter Chatter | fern, burt, pip, lowell, bleep |
| 🗣️ | sim | Babblish | lottie, ben, kiki, hank, whisper, bubs |
| 🎪 | vemmo | Bloopa Doopa | jesse, bram, mila |
| 🐾 | name | Name Talk | moki, grumbo, zee, lulu, bonk |
| 🗡️ | bark | Grunts & Barks | hero, heroine, elder, kid, brute |
| 👹 | grumble | Grumble | grok, moss, burly |
| 🤖 | gizmo | Gizmo | bolt, cog, unit9 |
| 🧚 | sprite | Sprite | fae, wisp, glimmer |
| 👻 | ghost | Spooky Whisper | whisper, wail, hollow |
| 🏘️ | echo | Echo Folk | village, crowd, kids |
| 📻 | radio | Mumble Radio | dispatch, pilot, scout |
| 👶 | baby | Babytalk | tot, bubba, mini |
| 🎐 | tone |
🛠️ Repository
site/: the tool as published on babble.wrongfloor.studio. Plain HTML and JavaScript, no build step. site/src/engine.js is the engine, shared by the tool and the web moduleaddons/babble_engine/: the Godot 4 addontests/: node tests/check.mjs (copy, links, headers) and node tests/engine.mjs (every voice renders, seeds repeat, WAV is valid). Or just npm test. With Godot installed, GODOT=<path to the console exe> node tests/godot.mjs compares the addon with the web moduledocs/art/: the pictures in this README, the Asset Library icon (icon.svg → icon.png) and the Godot demo screenshot
📜 License
Code: MIT, see LICENSE. Audio you make with Babble Engine is yours to use, also commercially. No credit needed, but a mention is always welcome 💛
