@mumble-speech/web

Video-game "mumble speech" for the web: turn a line of text into the babble that game characters speak, in three styles — beep (Undertale), grunt (Banjo-Kazooie) and alphabet (Animal Crossing).

WebAssembly build of the mumble-speech Rust core: same output, bit for bit, as the native, Godot and Unity bindings. Works in browsers and Node ≥ 20, with any bundler (the module is inlined).

Early development, not published to npm yet.

Usage

import { init, render, Speaker, WebAudioPlayer, encodeWav } from '@mumble-speech/web';

await init();

// A whole line: samples plus a timeline to sync text with sound.
const { samples, sampleRate, timeline } = render('Hello, world!', { style: 'alphabet', pitch: 2 });
const player = new WebAudioPlayer();
player.unlock(); // inside a click handler
await player.play(samples, sampleRate);

// Typewriter dialogue: one character at a time.
const speaker = new Speaker({ style: 'grunt', seed: 42 });
for (const ch of 'Guh-huh!') {
  const { samples, advance } = speaker.next(ch); // play `samples`, wait `advance` seconds
}
speaker.free();

// Download as WAV.
const wav = encodeWav(samples, sampleRate);

Voice: style, pitch (−12..12 semitones, default 0), speed (5..30 characters/s, default 15), variation (0..1, default 0), seed (default 0). Invalid values throw a MumbleError with a code.

Each timeline event has start/duration (seconds), utf16Start/utf16Length (index your JavaScript string with them) and sound: { style, variant }: which sound plays (grunt 0–3, alphabet 0–25 for a–z, 26 neutral, beep 0). Map it to whatever you like — mouth shapes, particles, colors. Nothing is drawn by the package.

License

MIT OR Apache-2.0.