Home Assistant voice announcements: speaking through the house with a Piper voice

When you speak to Assist the reply returns to that satellite or phone. An announcement instead goes to the speaker you chose, so 'the front door is open' comes out of the living-room speaker, not the device you spoke to.

Once that difference is clear, the two get designed separately: conversation near the satellite, announcements in the room where people are.

In an automation action, choose tts.speak, target the TTS entity (for example the one Piper provides), and give the speaker and the sentence as data. The speaker is a media_player entity.

Repeating the same sentence reuses the cache and is faster from the second time. Values such as time or temperature change every run and miss the cache, so keep fixed announcements apart from value-carrying ones.

With several speakers, name them by room and pick the target by where the event happened: the entry sensor to the living room, the washing machine to the laundry area.

To speak everywhere at once, group the speakers and send once. Sending to each room separately makes the sentences drift and overlap.

At night, lower the volume and postpone anything not urgent until morning. Only urgent events, such as a door opening or smoke, should make noise at night.

Volume has to be set per speaker. The same value sounds different on a small and a large speaker, so listen once in each room before fixing it.

A long sentence takes longer to synthesize and is easier to miss halfway. Keep announcements to two sentences and send the detail to a phone notification or a screen.

Numbers read more reliably written as they sound, 'twenty-three degrees' rather than '23'. Writing device names the way they should sound cuts odd readings.