wakewordlab

On-device wake word detection for Python.

Includes a Silero VAD pre-filter that gates inference on speech frames only, reducing CPU usage on silence and cutting false positives.

Installation

pip install wakewordlab          # file scoring only
pip install wakewordlab[mic]     # + microphone streaming

Quick start

import wakewordlab

# Download a model (cached to ~/.cache/wakewordlab/models/)
wakewordlab.download("hey_jarvis")

# Start listening — VAD is on by default
detector = wakewordlab.WakewordDetector("hey_jarvis")

@detector.on_detection
def handle(event):
    print(f"Detected: {event.wake_word}  confidence={event.confidence:.2f}")

detector.start()
detector.wait()   # blocks until Ctrl-C

Streaming with VAD

VAD (Silero v5) is enabled by default. It gates the wake word model so inference only runs on frames that contain speech, keeping CPU usage low during silence.

detector = wakewordlab.WakewordDetector(
    "hey_jarvis",
    vad=True,            # default — Silero VAD pre-filter
    vad_threshold=0.5,   # lower = more sensitive (pass more frames)
    threshold=0.5,       # wake word detection threshold
    cooldown_sec=1.5,    # min seconds between consecutive detections
    stride_sec=0.1,      # how often a window is scored (seconds)
)

@detector.on_detection
def handle(event):
    print(f"{event.wake_word}  confidence={event.confidence:.2f}")
    # event.audio — the 1s window that triggered (16 kHz float32 numpy array)

with detector:
    detector.wait()

To disable VAD (score every window regardless of speech):

detector = wakewordlab.WakewordDetector("hey_jarvis", vad=False)

Score an audio file

result = detector.score_file("clip.wav")
# {
#   "detected": True,
#   "peak_score": 0.92,
#   "peak_time_sec": 0.4,
#   "hit_count": 3,
#   "window_count": 21,
#   "threshold": 0.5,
# }

Score a raw array

import numpy as np
audio = np.zeros(16000, dtype=np.float32)   # 1 s at 16 kHz
prob = detector.score(audio)                 # float 0–1

List available models

wakewordlab.list_models()   # → ["hey_jarvis", ...]

Options

List audio input devices

from wakewordlab.audio import AudioStream
print(AudioStream.list_devices())

Commercial models

Commercial models are distributed as .wkw files with a license key:

detector = wakewordlab.WakewordDetector(
    "path/to/custom.wkw",
    license_key="xxxx-xxxx-xxxx-xxxx",
)

License

Public models are licensed for non-commercial use only.