Skip to content

Questions and answers

The device runs microWakeWord and openWakeWord models in TensorFlow Lite format: a .tflite file. Other formats, such as .onnx, don’t work; download the .tflite version.

Two community libraries have hundreds of ready-made wake words, and both let you train your own:

With the EchoLocal integration: open the EchoLocal dashboard in the sidebar, go to Wake words, and drop the .tflite file onto it. It’s offered to every device immediately, with no restart.

To add a model manually:

  1. In your Home Assistant config directory, create a folder named custom_wake_words.

  2. Copy the .tflite file into it.

  3. Next to it, add a .json file with the same name. If the model came with one, use that. Otherwise write one like this, with the phrase you’ll say and the model’s file name:

    {
    "type": "micro",
    "wake_word": "Hey Computer",
    "model": "hey_computer.tflite"
    }
  4. Restart Home Assistant. It only looks in custom_wake_words when it starts.

Pick it as an assistant’s Wake word in Home Assistant.

You can also pick it on an Echo Show’s screen. On the Echo Dot, use Home Assistant. See Voice assistants.