Methods, systems, and apparatus for voice authentication and command. In an aspect, a method comprises: receiving, by a data processing apparatus that is operating in a locked mode, audio data that encodes an utterance of a user, wherein the locked mode prevents the data processing apparatus from performing at least one action; providing, while the data processing apparatus is operating in the locked mode, the audio data to a voice biometric engine and a voice action engine; receiving, while the data processing apparatus is operating in the locked mode, an indication from the voice biometric engine that the user has been biometrically authenticated; and in response to receiving the indication, triggering the voice action engine to process a voice action that is associated with the utterance.