Python Wake Word Detection Guide

Step-by-step guide to adding on-device wake word detection to your Python application.

New to voice activation? Read what a wake word is and how wake word detection works before choosing an integration path.

Python Integration Guide

📦 Repository

GitHub: https://github.com/frymanofer/Python_WakeWordDetection

Python SDK for wake word detection with NumPy and ONNX Runtime.

Step 1: Installation

# Install from PyPI
pip install davoice-wake-word

# Or install from source
git clone https://github.com/frymanofer/Python_WakeWordDetection
cd python-wake-word
pip install -e .

Step 2: Basic Usage

from davoice import KeywordDetector
import sounddevice as sd

def on_detection(result):
    print(f"Wake word detected!")
    print(f"Confidence: {result['confidence']:.2f}")
    print(f"Timestamp: {result['timestamp']}")

# Initialize detector
detector = KeywordDetector(
    model_path="models/hey_jarvis.onnx",
    threshold=0.97,
    sample_rate=16000
)

# Start detection
print("Listening for wake word...")
detector.start(callback=on_detection)

# Keep running
try:
    while True:
        sd.sleep(1000)
except KeyboardInterrupt:
    detector.stop()
    print("Stopped listening")

Step 3: Advanced Configuration

from davoice import KeywordDetector, DetectorConfig
import numpy as np

# Advanced configuration
config = DetectorConfig(
    model_path="models/custom_wake_word.onnx",
    threshold=0.95,
    buffer_count=3,
    min_detection_interval=2.0,  # seconds
    enable_noise_reduction=True,
    log_level="INFO"
)

detector = KeywordDetector(config)

# Custom audio processing
def audio_callback(indata, frames, time, status):
    if status:
        print(f"Audio status: {status}")

    # Process audio chunk
    audio_data = np.frombuffer(indata, dtype=np.float32)
    result = detector.process_audio(audio_data)

    if result and result['detected']:
        print(f"Wake word detected! Confidence: {result['confidence']:.3f}")

# Start streaming with custom callback
import sounddevice as sd

stream = sd.InputStream(
    samplerate=16000,
    channels=1,
    callback=audio_callback,
    blocksize=1024
)

with stream:
    print("Listening...")
    sd.sleep(60000)  # Run for 60 seconds

Step 4: Multi-Model Detection

from davoice import MultiModelDetector

# Detect multiple wake words simultaneously
detector = MultiModelDetector([
    {
        'name': 'hey_jarvis',
        'model_path': 'models/hey_jarvis.onnx',
        'threshold': 0.97
    },
    {
        'name': 'ok_computer',
        'model_path': 'models/ok_computer.onnx',
        'threshold': 0.95
    }
])

def on_multi_detection(result):
    print(f"Detected: {result['model_name']}")
    print(f"Confidence: {result['confidence']:.2f}")

detector.start(callback=on_multi_detection)

# Or check manually
while True:
    results = detector.get_latest_detections()
    for result in results:
        if result['detected']:
            print(f"{result['name']}: {result['confidence']:.2f}")

✅ Use Cases

  • • Desktop applications and voice assistants
  • • Raspberry Pi and IoT devices
  • • Server-side voice processing
  • • Research and prototyping

Need Help Getting Started?

Our team provides free integration support and custom wake word training for all platforms.

Free Development Support • Code Examples • Community Discord • Custom Wake Words in 1–2 Working Days