react-native-sherpa-onnx-offline-tts

react-native-sherpa-onnx-offline-tts

A lightweight React Native wrapper around Sherpa‑ONNX that lets you run 100 % offline Text‑to‑Speech on iOS and Android.

✨ Features

🔊 Offline – all synthesis happens on‑device, no network needed	⚡ Fast – real‑time (or faster) generation on modern phones
🎙️ Natural voices – drop‑in support for Piper / VITS ONNX models	🛠️ Simple API – a handful of async methods you already know

📦 Installation

# Add the library
npm install react-native-sherpa-onnx-offline-tts
# or
yarn add react-native-sherpa-onnx-offline-tts

# iOS only\	npx pod-install

Minimum versions | Android 5.0 (API 21) • iOS 11

🚀 Quick Start

• Choose a model – grab any Piper voice ZIP (e.g. vits-piper-en_US-ryan-medium.zip) and host it yourself or bundle it with the app.
• Download & unzip the archive into your app’s sandbox (the example below uses react‑native‑fs & react‑native‑zip‑archive).
• Create a config JSON with absolute paths to *.onnx, tokens.txt, and the espeak-ng-data folder.
• Initialize, then generate or stream speech.

import TTSManager from 'react-native-sherpa-onnx-offline-tts';
import RNFS from 'react-native-fs';
import { unzip } from 'react-native-zip-archive';

const MODEL_URL =
  'https://example.com/vits-piper-en_US-ryan-medium.zip';

async function setupTTS() {
  const archive = `${RNFS.DocumentDirectoryPath}/vits.zip`;
  const extractRoot = `${RNFS.DocumentDirectoryPath}/extracted`;

  // 1️⃣  Download if missing
  if (!(await RNFS.exists(archive))) {
    await RNFS.downloadFile({ fromUrl: MODEL_URL, toFile: archive }).promise;
  }

  // 2️⃣  Unpack if first run
  if (!(await RNFS.exists(`${extractRoot}/vits-piper-en_US-ryan-medium`))) {
    await unzip(archive, extractRoot);
  }

  // 3️⃣  Point the engine to the files
  const base = `${extractRoot}/vits-piper-en_US-ryan-medium`;
  const cfg = {
    modelPath: `${base}/en_US-ryan-medium.onnx`,
    tokensPath: `${base}/tokens.txt`,
    dataDirPath: `${base}/espeak-ng-data`,
  };

  // 4️⃣  Initialise (only once per session)
  await TTSManager.initialize(JSON.stringify(cfg));
}

async function sayHello() {
  const text = 'Hello world – spoken entirely offline!';
  const speakerId = 0;   // Piper uses 0 for single‑speaker models
  const speed = 1.0;     // 1 == default, < 1 slower, > 1 faster

  await TTSManager.generateAndPlay(text, speakerId, speed);
}

📚 API Reference

Method	Signature	Description
initialize	(modelConfigJson: string): Promise<void>	Must be called once before any synthesis. Pass a JSON string with modelPath, tokensPath, dataDirPath.
generate	(text: string, speakerId: number, speed: number): Promise<string>	Generates speech and returns the path of the WAV file.
generateAndPlay	(text: string, speakerId: number, speed: number): Promise<void>	Generates speech and streams it to the device speaker.
stopPlaying	(): void	Immediately stops playback.
addVolumeListener	(cb: (volume: number) => void): EmitterSubscription	Subscribes to real‑time RMS volume callbacks during playback. Call subscription.remove() to unsubscribe.
deinitialize	(): void	Frees native resources – call this when your app unmounts or goes to background for a long time.

🔊 Supported Models

• Any Piper VITS model (*.onnx) with matching tokens.txt and espeak-ng-data directory.
• Multi‑speaker models are supported – just pass the desired speakerId.

Need other formats? Feel free to open an issue or pull request.

🛠️ Example App

A minimal, production‑ready example (downloads the model on first launch, shows a progress spinner, animates to mic volume, etc.) lives in example/App.tsx – the snippet below is an abridged version:

const App = () => {
  /* full source lives in the repo */
  return (
    <View style={styles.container}>
      {isDownloading ? (
        <ProgressBar progress={downloadProgress} />
      ) : (
        <>
          <AnimatedCircle scale={volume} />
          <Button title="Play" onPress={handlePlay} disabled={isPlaying} />
          <Button title="Stop" onPress={handleStop} disabled={!isPlaying} />
        </>
      )}
    </View>
  );
};

🤝 Contributing

Bug reports and PRs are welcome! Please see CONTRIBUTING.md for the full development workflow.

📄 License

MIT

Made with ❤️ & create‑react‑native‑library