Home voices: family or guest
Homer recognizes who is speaking. The owner marks the household's voices; everyone else is a guest to the speaker: lights, music and chat stay open to them, while cameras, tasks, memory and plan editing are off.
The voice album is one per home and lives on the Homer server. Whichever speaker hears the phrase — the one on the computer, the backup or a satellite in another room — it is matched against the same album, so a person recognized at one speaker is recognized at the others too.
How to add your voice
The easiest way is by voice:
| Say | What happens |
|---|---|
| “Homer, remember my voice, I am Alex” | “Got your voice, Alex. The owner will confirm it in the app; meanwhile keep talking to me so I learn it better”. A request appears in “Household voices” |
| “Homer, remember my voice” | The same without a name — the owner fills it in |
| The same when the voice is already household | “I already know your voice, Alex” |
“Learn my voice”, “my name is Alex” and “I'm Alex” work too. If the speaker answers “I couldn't catch the voice. Say it again, a bit longer”, the phrase was too short.
A request is not trust yet: a voice becomes household only after the owner confirms it.
“Household voices”
Settings → Speaker → Household voices, in the web and in the app. Visible only to the owner.
- Requests: voices that asked to be remembered. Confirm makes the voice household, Reject leaves it a guest.
- Household: confirmed voices. Each has a name and a role: “household” or “owner”. Save name after editing, Forget deletes the voice with its prints.
- Other voices (N): everyone the speakers have heard, that is, guests. Each shows its last phrases so you can tell who it is, the phrase count and the marks “male”, “female”, “child?”. Household makes the voice household; “Same person as…” plus Merge joins two voices of one person — for example, recorded at different speakers.
- Until someone talks to Homer, the list is empty: voices appear by themselves from phrases addressed to Homer.
A print is taken only from phrases addressed to Homer: other people's conversations, the TV and the speaker's own echo never enter the album. A print is not a recording; speech can't be restored from it.
What you see
- In Speaker conversations, each phrase is prefixed with who said it: a name or “guest”.
- In Home Assistant, each speaker has a “Speaking” sensor,
sensor.homer_speaking_<speaker>: the name of the last person who spoke, with the role and confidence as attributes. Good for automations: a night light in the kids' room by the child's voice. - The brains know who is talking (“Speaking: Alex, owner”) and can answer by name.
- If two speakers heard the phrase, a recognized household voice is an argument in the speaker contest: the speaker that recognized it more confidently answers — it is closer.
What a guest can't do
As long as the home has at least one confirmed household voice, a guest by voice can't use:
- cameras: showing or checking what a camera sees;
- tasks for the bigger model;
- Homer's long-term memory;
- home plan editing.
The speaker refuses in one sentence and suggests asking the household. Lights, outlets, music, timers, questions and chat are open to everyone. While there are no household voices, there are no restrictions: everyone is equal, as before.
A voice is not a password
A recording or the TV can fake a voice. So a role by voice only narrows rights and never opens anything: the task PIN stays, and text from Telegram, the app or the Alice skill doesn't count as a guest — there is no voice there.
What you need
The voice recognition models come from Yandex speaker firmware, and Homer doesn't distribute them: put them into the speaker's speaker-home\voiceid, as described in Speaker on the computer. Without the models the speaker works as before, and “Household voices” stays empty. Where the files come from and on what terms: Models and licenses.