It is available via Settings > Voice Assistant and allows to enter commands via voice.
Add assistant gives it a name and configures it as for the languages spoken. It appears then in the menu and can be set as being the preferred.
There is also the offer to get the cloud service that also offers connection to Google Assistant and Amazon Alexa.
For first tests just the written sentences must be understood, so language to speech and speech to language can be left empty.
In settings > Developer Tools > Assist is a Sentence parser, Enter a command like: Go to 3 a YAML file is created but HomeAssistant is juts guessing what to do with this voice command and this is probably not want is desired
In automations the trigger "when a sentence is said" can be set. And for test the action "Set conversation response" with a text (without {{....}} template).
Triggers can understand sentences with variables as: Go to {floor} and "Set conversation response" with "sent to {{ trigger.slots.floor }}"
Instead a "Set conversation response" action a "MQTT publish" action can be defined with its topic and payload {{ trigger.slots.floor }}
Linux Voice Assistant LVA https://github.com/OHF-Voice/linux-voice-assistant (replaced wyoming) is the listener and passing forward part it supports also wake-words and obviously deals with the microphone and speaker.