Voice Labs experiments with voice commands shaped around the people and rooms in a home. Households can define phrases that run a script, activate a scene, carry out an action or speak a response. Commands can apply throughout the home or only in one room, so the same words can do something different depending on which satellite hears them.
The project is experimental and unsupported, and currently hears and speaks English. Trying it requires a Home Assistant app, a custom integration installed through HACS and alternate firmware for Voice Preview Edition satellites. Michael Hansen makes his first contribution to the experimental-apps repository with the addition.
Command matching runs on the Home Assistant host, but the speech-engine settings determine where other audio processing happens. If Home Assistant Cloud is signed in and no engine has been chosen, the app initially selects its speech recognition and voice, streaming every command as it is spoken. Owners can instead choose the app’s built-in models and keep that processing local.