Naming your assistant, and choosing its voice and face
Everything on the Voice & appearance page: what your mirror is called, the wake word, the seven voices it can speak in, the four personalities, and the ten faces it can wear.
Where these settings live
Choose Assistant in the navigation and you land on Voice & appearance. Everything on this page belongs to the household rather than to one screen: there is one name, one voice, one personality and one face across every mirror you own.
The page saves as you go. There is no Save button — a change writes itself and a small "Saving…" then "Saved ✓" appears at the top. Voice, name and personality take effect on the next thing anybody says to a screen, because the assistant re-reads its identity at the start of every turn. A change of face is pushed to every paired screen straight away, without anybody reloading anything.
Every setting on this page
The four personalities
The personality is added to what the assistant is told about itself at the start of every turn, so a change applies to the very next thing said. It does not survive into old transcripts — those are what was actually said at the time.
Choosing a voice
Seven voices are offered. All of them are synthesised with a neural engine, which is why they breathe and stress words rather than reading them off a card.
Press ▶ Hear beside a voice
It speaks a sample line using your assistant's name: "Good evening. I'm Mirra. You have three things on today, and it should stay dry." You can audition without pairing a screen first.
Press Use to choose it
The row you have chosen is highlighted and the Use button disappears from it. That is the whole of the change — the next thing any screen says is in the new voice.
- Amy — British. The default.
- Emma — British.
- Brian — British.
- Arthur — British.
- Joanna — American.
- Matthew — American.
- Olivia — Australian.
Previews are limited to thirty in five minutes per person, because each one is real synthesis and costs real money. If you hit that you will be asked to slow down for a moment. The replies themselves are unaffected.
The wake word, and what it actually does
The wake word is a stored setting and nothing reads it today. It is kept for the Raspberry Pi sense agent — the daemon intended to run beside the kiosk browser on a Pi and own the microphone, the camera and everything else a browser handles badly — and that agent is a skeleton. It serves a socket and emits a heartbeat; wake-word detection is not written, and the screen app does not connect to it.
On a screen that is a browser, the wake word does nothing at all. There is no listening in the background and no phrase that summons anything. You summon the assistant with the microphone button, with M, or with T to type. The page says as much beneath the field, and it is worth reading it literally rather than hopefully.
Setting a wake word does not make any screen listen for it — not today, on any hardware. If you have set one and nothing happens when you say it, that is the expected behaviour and not a fault.
The ten faces
Under "Appearance when summoned" is a grid of tiles. Each tile is the real face running, not a screenshot, so what you see there is what the mirror will do. Tiles start when they scroll into view, so the lower ones take a moment to wake up. The one you are using is marked Current.
Every face lip-syncs to the speech — the mouth shapes come from the same synthesis that produces the audio, so even the faces without a mouth move in time with the words.
Tuning a face
Press Preview & tune on a tile
Faces with dials say "Preview & tune"; the rest say "Preview". A large preview opens with the sliders beside it.
Move a slider and watch
Most dials feed straight into the running preview. The two that change the shape of the thing — thread density and tentacle count — rebuild the preview instead, so those blink.
Press Use this face, or Done
If you are tuning the face you already use, the dials save as you move them. If you are tuning a different one, they are only a preview until you press "Use this face" — which selects it and keeps the numbers together. Done closes the panel; so does Escape.
The Twinview Legacy preview sweeps the whole colour range so you can see all of it. On the wall it will sit wherever today's forecast has put it, which for most of a British winter means blue.
What renaming does not change
The name reaches the assistant itself — it is told what it is called, so it uses that name when it talks about itself, and the voice previews speak it. It does not rename the product around it. The typed-input box on a screen still reads "Ask Mirra…", the portal navigation still says Assistant, and the transcripts still label the other side Mirra.
There is also no per-person voice and no per-screen face. One household, one identity. If you want the bedroom mirror to behave differently from the kitchen one, that is a job for scenes and the scheduler rather than for this page.
Related
- The assistant: how a conversation worksHow you summon Mirra, what happens between hearing you and answering you, which tools it can reach, where the transcript goes, and what a conversation costs against your monthly allowance.
- What Mirra remembers, and how to see or delete itMirra stores only the facts somebody asked it out loud to remember, shows every one of them in the words that produced them, and can be told to stop remembering altogether.
- Letting the assistant look things upHow to switch web search on for your assistant, what a search costs, and why a conversation that reads the web can no longer change anything in your home.
- What the assistant will and will not doThe limits built into Mirra's assistant: locks off by default, quiet hours, the greeting cooldown, and the requests it refuses outright.
- GreetingsHow Mirra greets people — the greeting widget on the glass, the spoken greeting when a screen recognises somebody, and the rules, quiet hours and cooldown that decide what is said and when.