🎙️ LipSync.js

High-quality lip-sync animation from audio in the browser

WebAssembly port of Rhubarb Lip Sync with TypeScript support.

✨ Features

• 🚀 High Performance - Runs natively in browser via WebAssembly
• 🎯 Accurate - Uses PocketSphinx speech recognition for precise phoneme detection
• 📦 Small Bundle - Only ~80KB JavaScript + 2.2MB WASM + models
• 🔧 TypeScript - Full type definitions included
• 🌐 Framework-Agnostic - Works with React, Vue, Svelte, vanilla JS, and any framework
• 🧵 Web Workers - Non-blocking analysis with worker pool
• 🎨 Complete API - Audio utilities, format conversion, microphone recording
• 📱 Browser-Native - No server required, runs entirely client-side

📦 Installation

npm install lip-sync-js

🚀 Quick Start

Vanilla JavaScript / TypeScript

import { analyze, recordAudio } from 'lip-sync-js';

// Record audio from microphone
const { pcm16 } = await recordAudio(5000); // 5 seconds

// Analyze
const result = await analyze(pcm16, {
  dialogText: "Hello world", // Optional, improves accuracy
  sampleRate: 16000
});

// Use mouth cues for animation
result.mouthCues.forEach(cue => {
  console.log(`${cue.start}s - ${cue.end}s: ${cue.value}`);
  // Output: 0.00s - 0.35s: X
  //         0.35s - 0.50s: D
  //         0.50s - 0.85s: B
  //         ...
});

React

import { useState, useEffect, useRef } from 'react';
import { LipSync, recordAudio } from 'lip-sync-js';

function useLipSync() {
  const [result, setResult] = useState(null);
  const lipSyncRef = useRef(LipSync.getInstance());

  useEffect(() => {
    lipSyncRef.current.init();
    return () => lipSyncRef.current.destroy();
  }, []);

  const analyze = async (pcm16, options) => {
    const result = await lipSyncRef.current.analyze(pcm16, options);
    setResult(result);
  };

  return { analyze, result };
}

function MyComponent() {
  const { analyze, result } = useLipSync();

  const handleRecord = async () => {
    const { pcm16 } = await recordAudio(5000);
    await analyze(pcm16, { dialogText: "Hello world" });
  };

  return (
    <div>
      <button onClick={handleRecord}>Record & Analyze</button>
      {result && <div>Found {result.mouthCues.length} mouth cues!</div>}
    </div>
  );
}

See examples/react for complete example.

Vue

<script setup lang="ts">
import { ref, onMounted, onUnmounted } from 'vue';
import { LipSync, recordAudio } from 'lip-sync-js';

const result = ref(null);
const lipSync = LipSync.getInstance();

onMounted(() => lipSync.init());
onUnmounted(() => lipSync.destroy());

const handleRecord = async () => {
  const { pcm16 } = await recordAudio(5000);
  result.value = await lipSync.analyze(pcm16, { dialogText: "Hello world" });
};
</script>

<template>
  <div>
    <button @click="handleRecord">Record & Analyze</button>
    <div v-if="result">Found {{ result.mouthCues.length }} mouth cues!</div>
  </div>
</template>

See examples/vue for complete example.

Svelte

<script>
import { writable } from 'svelte/store';
import { LipSync, recordAudio } from 'lip-sync-js';

const result = writable(null);
const lipSync = LipSync.getInstance();
lipSync.init();

async function handleRecord() {
  const { pcm16 } = await recordAudio(5000);
  const res = await lipSync.analyze(pcm16, { dialogText: "Hello world" });
  result.set(res);
}
</script>

<button on:click={handleRecord}>Record & Analyze</button>
{#if $result}
  <div>Found {$result.mouthCues.length} mouth cues!</div>
{/if}

See examples/svelte for complete example.

📚 Documentation

• Getting Started
• API Reference
• Framework Examples:
  • Vanilla JS
  • React
  • Vue
  • Svelte

🎨 Mouth Shapes (Visemes)

LipSync.js generates 9 mouth shapes based on Preston Blair's phoneme categorization:

🔬 How It Works

1. Speech Recognition - PocketSphinx analyzes audio to detect phonemes
2. Phoneme Mapping - Phonemes are mapped to Preston Blair mouth shapes
3. Timing Optimization - Animation is smoothed for natural transitions
4. JSON Output - Returns timestamped mouth shape cues

📊 API Overview

Core Functions

// Simple one-off analysis
import { analyze } from 'lip-sync-js';
const result = await analyze(pcm16, options);

// Async analysis (non-blocking)
import { analyzeAsync } from 'lip-sync-js';
const result = await analyzeAsync(pcm16, options);

// Using the main class
import { LipSync } from 'lip-sync-js';
const lipSync = LipSync.getInstance();
await lipSync.init();
const result = await lipSync.analyze(pcm16, options);

Audio Utilities

import {
  recordAudio,
  loadAudio,
  audioBufferToInt16,
  float32ToInt16,
  resample
} from 'lip-sync-js';

// Record from microphone
const { pcm16, audioBuffer } = await recordAudio(5000); // 5 seconds

// Load from file or URL
const { pcm16, audioBuffer } = await loadAudio('audio.mp3');

// Convert formats
const int16 = audioBufferToInt16(audioBuffer, 16000);
const int16 = float32ToInt16(float32Array);
const resampled = resample(float32Array, 44100, 16000);

Types

interface MouthCue {
  start: number;  // seconds
  end: number;    // seconds
  value: string;  // X, A, B, C, D, E, F, G, or H
}

interface LipSyncResult {
  mouthCues: MouthCue[];
  metadata?: {
    duration: number;
    sampleRate: number;
    dialogText?: string;
  };
}

interface LipSyncOptions {
  dialogText?: string;  // Improves accuracy significantly
  sampleRate?: number;  // Default: 16000 (recommended)
}

🛠️ Development

# Install dependencies
npm install

# Build WASM module
npm run build:wasm

# Build TypeScript
npm run build:ts

# Build everything
npm run build

# Type check
npm run typecheck

# Clean build artifacts
npm run clean

📄 License

MIT License - see LICENSE

🙏 Credits

• Original Rhubarb Lip Sync by Daniel Wolf
• PocketSphinx for speech recognition
• Preston Blair for phoneme categorization system

🐛 Issues

Report issues at https://github.com/biolimbo/lip-sync-js/issues

📈 Roadmap

• Full Web Worker implementation
• Multiple language support
• Custom phoneme-to-viseme mappings
• Real-time streaming analysis
• Three.js / Babylon.js helpers
• Unity WebGL bridge