ONNX BABY 🤘🔥

ROCK. ROLL. BABY.

Ultra-simple Text-to-Speech built with pure TypeScript and ONNX Runtime. Just load a .onnx model and let it scream.

🎤 What is ONNX BABY?

ONNX BABY is a lightweight TTS runtime for Node.js that:

• Loads ONNX voice models
• Generates PCM audio
• Streams audio in real time
• Runs fully in TypeScript

It's designed for:

• Real-time servers
• AI assistants
• Games
• Voice bots
• Experimental audio projects

🚀 Features

• ⚡ Pure TypeScript
• 🧠 ONNX Runtime (Node binding)
• 🔊 PCM streaming
• 📦 Zero native compilation
• 🎛 Simple API
• 🤘 Fast startup

�� Installation

npm install onnx-baby

🧠 Load a Voice

import { setVoice } from "onnx-baby";

await setVoice("./models/model.onnx");

Model loads once into memory.

🔊 Generate Audio (Streaming)

import { streamTextToAudio } from "onnx-baby";

const stream = streamTextToAudio("ROCK AND ROLL BABY!", {
  sentenceSilence: 0.3,
  expressiveness: 0.4,
});

for await (const chunk of stream) {
  console.log("Chunk size:", chunk.length);
}

Output format:

• PCM 16-bit
• Mono
• 22050 Hz (default)

🌐 Example: Express + WebSocket

import express from "express";
import { WebSocketServer } from "ws";
import { setVoice, streamTextToAudio } from "onnx-baby";

await setVoice("./models/model.onnx");

const app = express();
const server = app.listen(3000);

const wss = new WebSocketServer({ server });

wss.on("connection", (ws) => {
  ws.on("message", async (msg) => {
    const { text } = JSON.parse(msg.toString());

    const stream = streamTextToAudio(text);

    for await (const chunk of stream) {
      ws.send(chunk);
    }

    ws.send(JSON.stringify({ done: true }));
  });
});

⚙️ Options

| Option | Type | Default | Description | | --------------- | ------ | ------- | ------------------------- | | sentenceSilence | number | 0.3 | Pause between sentences | | expressiveness | number | 0.2 | Emotion intensity | | speed | number | 1.0 | Speech rate multiplier |

🧪 Performance

• Model load: ~100–400ms (depends on size)
• First audio chunk: typically < 150ms
• Memory footprint: minimal (single model instance)

🧱 Architecture

Text
 ↓
Tokenizer
 ↓
ONNX Runtime (Node)
 ↓
PCM Buffer
 ↓
Stream (Async Generator)

📌 Use Cases

• Real-time AI voice agents
• Browser TTS via WebSocket
• MMO NPC voice systems
• Streaming voice assistants
• CLI voice tools

🛣️ Roadmap

• [ ] Built-in WAV encoder
• [ ] Multi-model hot switching
• [ ] Voice pitch control
• [ ] Browser WASM build
• [ ] GPU acceleration support

📜 License

MIT

🤘 Philosophy

ROCK. ROLL. BABY.