WakeCore sample hotwords
Open hotword detection for private voice systems. Lokale Hotword-Erkennung für private KI-Systeme.
Eight ready-made .wake files for the WakeCore SDK.
Detection runs entirely on your device: no cloud calls, no API keys, no telemetry.
Requirements (as of October 2026)
.wakeformat v2 (regenerated on 2026-10-28). The files published here in May 2026 used format v1, which the current engine rejects (unsupported or damaged .wake file).- The WakeCore engine for Linux x86_64 (glibc 2.29 or newer), as published on
2026-10-28 (
wc_version()reports0.1.0). Other platforms are not available yet. - SDK 0.1.1 or newer recommended (0.1.0 ignores the
sensitivityargument).
Quick start (Python)
pip install "wakecore[audio]"
wakecore install-engine # engine -> ~/.wakecore/0.1.0/, SHA-256 checked
hf download keyvan-ai/wakecore --local-dir wakecore-hotwords
wakecore check wakecore-hotwords/hotwords/hey_computer.wake
wakecore listen wakecore-hotwords/hotwords/hey_computer.wake
from wakecore import Runtime, AudioCapture
with Runtime.load("hotwords/hey_computer.wake") as rt, \
AudioCapture(rt.sample_rate, rt.frame_length) as mic:
for frame in mic: # 512 samples, 16 kHz, mono, int16
if rt.process(frame):
print("wake!")
Node.js (npm install wakecore), C, C++, Rust and C# use the same files and
engine; see the GitHub README.
Files
| File | Say | Default sensitivity | Detected | False alarms in speech | Fires on other samples |
|---|---|---|---|---|---|
hotwords/computer_an.wake |
Computer an | 0.4 | 34/36 | 8 (all with "Computer") | 9/12 Hey Computer |
hotwords/guten_morgen.wake |
Guten Morgen | 0.7 | 24/36 | 1 | – |
hotwords/hallo_welt.wake |
Hallo Welt | 0.7 | 33/36 | – | – |
hotwords/hey_assistant.wake |
Hey Assistant | 0.4 | 35/36 | 2 ("the assistant …") | – |
hotwords/hey_computer.wake |
Hey Computer | 0.4 | 36/36 | 6 (all with "Computer") | 11/12 Computer an |
hotwords/hey_keyvan.wake |
Hey Keyvan | 0.5 | 36/36 | – | – |
hotwords/okay_deutsch.wake |
Okay Deutsch | 0.5 | 36/36 | – | – |
hotwords/wake_up.wake |
Wake up | 0.5 | 35/36 | 1 ("woke up") | – |
Measured on 2026-10-28 at each file's default sensitivity with three synthetic voices: Detected = 12 recordings of the phrase × three conditions (normal, 20 dB quieter, room reverb). False alarms in speech = recordings with a detection in 9.3 minutes of German and English speech without the hotword (21 of 48 texts contain similar words such as "Computer" or "woke up"). Fires on other samples = recordings of another sample phrase that trigger the file. Computer an and Hey Computer react to the word "Computer" on its own and to each other, so use only one of them. Guten Morgen is the weakest file with reverb (4 of 12). Synthetic voices are not real rooms; sensitivity runs from 0 to 1 (higher = more detections and more false alarms), so test with your own microphone before relying on a value.
Audio format
16 kHz, mono, 16-bit signed little-endian PCM, 512 samples (32 ms) per frame.
License
The SDK and these sample files are published under the MIT license. The .wake
files contain a third-party component under the Apache License 2.0; see
hotwords/NOTICE and hotwords/LICENSE-Apache-2.0.txt and keep both with the
files when you pass them on.
Links
- GitHub: https://github.com/Keyvanhardani/wakecore
- Website and custom hotwords: https://wakecore.de
- PyPI: https://pypi.org/project/wakecore/ · npm: https://www.npmjs.com/package/wakecore