LumiaWakeWord

A wake word detection library using Vosk and SoX for real-time voice recognition.

Features

• Real-time wake word detection
• Multi-platform support (Windows, macOS, Linux)
• Audio device selection (especially useful for Windows)
• Confidence threshold filtering
• Dynamic grammar updates

Installation

npm install @lumiastream/wakeword

Quick Start

import { startWakeWord, listAudioDevices } from "@lumiastream/wakeword";

// List available audio devices
const devices = await listAudioDevices();
console.log("Available devices:", devices);

// Start wake word detection
const wakeWord = startWakeWord({
  grammar: ["hello", "lumia", "computer"],
  confidence: 0.7,
  device: "0" // Optional: specify audio device
});

// Handle detections
wakeWord.stdout.on("data", (data) => {
  const lines = data.toString().split("\n");
  for (const line of lines) {
    if (line.startsWith("voice|")) {
      const word = line.split("|")[1];
      console.log(`Wake word detected: ${word}`);
    }
  }
});

Audio Device Selection

Windows Users

Windows users often need to select the correct audio input device. Use these commands to find and test devices:

# List all available audio devices
npm run list-devices

# Interactive device testing
npm run test-devices

# Use a specific device
AUDIO_DEVICE=1 npm start

Setting the Audio Device

Method 1: Environment Variable

set AUDIO_DEVICE=1
npm start

Method 2: Programmatically

startWakeWord({
  device: "1",
  grammar: ["hello", "lumia"]
});

See AUDIO_DEVICES.md for detailed device configuration guide.

API Reference

listAudioDevices()

Returns a Promise that resolves to an array of available audio devices.

const devices = await listAudioDevices();
// Returns: [{ id: "0", name: "Microphone (Realtek)" }, ...]

startWakeWord(options)

Starts the wake word detection process.

Options:

• device (string): Audio device ID to use
• soxPath (string): Path to SoX binary (optional)
• sampleRate (number): Sample rate, default 16000
• grammar (string[]): Array of wake words to detect
• confidence (number): Confidence threshold (0-1), default 0.7

Returns: ChildProcess instance

Scripts

npm run list-devices    # List available audio devices
npm run test-devices    # Interactive device testing
npm start              # Start wake word detection

Example

Run the included example:

node example.js

Dependencies

• Vosk - Speech recognition toolkit
• SoX - Sound processing tool
• Vosk model included: vosk-model-small-en-us-0.15

Platform Notes

Windows

• Default uses device "0" if not specified
• Use npm run test-devices to find the correct device
• USB microphones may appear as separate devices

macOS/Linux

• Uses system default audio input automatically
• Device selection usually not needed

Troubleshooting

1. No audio detected on Windows:

   • Run npm run test-devices to find the correct device
   • Set AUDIO_DEVICE environment variable

2. Poor recognition quality:

   • Adjust confidence threshold (lower = more sensitive)
   • Try different audio devices
   • Ensure microphone is not muted

3. Device not found:

   • Ensure microphone is connected before starting
   • Check system audio settings

SoX Binaries

Pre-compiled SoX binaries are included for all platforms. Source: https://github.com/zackees/static-sox/tree/main/bin

License

See LICENSE file for details.