← All articles

How Do I Train a Custom Wake Word for macOS?

Create a custom offline wake-word model for macOS apps, local automations, and privacy-focused voice interfaces.

How to train

  1. Enter one or multiple keyword phrases.
  2. Select English, Chinese, Spanish, Japanese, German, French, or Hindi.
  3. Train and test, then optionally upload authorized recordings to improve real-world robustness.
  4. Download ONNX or TFLite and integrate local inference into your macOS app.

Multiple keywords

One model can recognize multiple keyword labels, so a macOS utility can map phrases such as start, pause, and stop to different actions.

macOS deployment notes

Use ONNX Runtime or TFLite Runtime with Core Audio, request microphone permission, and test built-in and external microphones. CPU and threshold behavior can differ between Intel and Apple silicon systems, so validate the packaged application on the Macs you support.

Train a custom wake word online →

View the open-source inference engine and demo models →
ONNX / TFLite · ESP32 · Raspberry Pi · Android · Web · Windows · macOS · Linux · Home Assistant