The speaker
A person in the page who presents it in a real voice.
Ask Claude: "Put a speaker on this page."
Then say four things:
- still or moving
- the size — face · half bust · bust · full body
- the person — an existing speaker (Valentina, Zalia, Grace) · a photo you give · or describe the person you want and the image is generated
- the voice — an existing voice · one picked from the voice library (say the language, the tone, man or woman) · or a clone of a real person, with their consent, from 30 seconds of clean speech
Claude builds it with the site-speaker recipe.
The two speakers
Still — she speaks but does not move. What Zalia does on the Comores leaflets and Grace in Lesotho: a photo and a recorded voice. Quick and almost free — only the voice costs.
Moving — the new capability: she speaks and her face moves with the words. Valentina. One master clip (1.20 EUR, once) plus centimes per part.
Still speaker
Speaks, does not move. Zalia · Grace · the AZFA circle.
Moving speaker
Speaks and moves. Valentina, live.
The sizes (for the moving speaker)
How much of the person the page shows. Every size speaks.
Full body
To the feet. The image is ready; the speaking version not yet.
Buttons and features
Her card offers one button. After the click she presents the page part by part, and two icons remain: pause, stop.
A real conversation with the visitor. We do not know how to do it well yet; its button stays off the page until we do. Where we stand, exactly: Interaction.
Interaction — talking with the speaker
How it works. The visitor asks, typed or spoken; an agent finds the answer in a small set of fact files we wrote, and speaks it back in the speaker's own voice. She answers only from those files — nothing invented, internal notes can never leak. This grounding is called RAG; our method is the voice-agent skill.
The reference. Text and voice, in French, on the Comores SARL page. 5 fact files + a conduct charter; tested 9 for 9, trick questions included.
Her agent, voice, pronunciation and 13 fact files are all wired; the buttons were withdrawn 02-08-2026. No visitor can talk with her yet.
Narration only, by verdict of 19-07-2026. Her legally verified leaflets wait as the knowledge; wake her when a public assistant is wanted.
What is missing, for Valentina. A full voice test on the live page (a page opened from disk cannot use the microphone) · cutting her off mid-answer, never verified · the greeting is long in voice mode · the face does not answer — during a conversation the bust steps aside and only the voice replies, which is the real meaning of "not ready" · switching text to voice loses the dialogue · cost per minute unmeasured and the agent has no domain lock. Full detail: the diagnosis.
| Track | Next task | Effort | It unlocks | |
|---|---|---|---|---|
| 1 | Prove Valentina's conversation | Run the full voice round trip on the live page, fix what fails, restore the two buttons for Frank's verdict | ~1 day | The first speaker a visitor can talk with |
| 2 | Guard every speaker's answers | Write Valentina's witness questions; give the existing guard a schedule that runs | Hours | Answers that stay right as facts change |
| 3 | One-click knowledge refresh | Build the fact editor (edit, press Publish); reuse for every speaker | 1–2 days | Colleagues keep the knowledge current |
| 4 | The face that answers | Test real-time mouth movement on a rented GPU against live audio | Days, exploratory | The speaker herself visibly answering |
| 5 | Wake Grace | When wanted: run the method on her leaflet corpus | ~1 day | Lesotho citizens ask instead of read |
| 6 | The sovereign bet | A bet, not a task: NVIDIA's self-hosted voice model — natural interruption, under a third of a second — English-only today; we asked about Spanish | Watch | Conversation on our own machines, at centimes |
How one is made
- Portrait on green — one photo, head and shoulders, so she can be cut out.
- Voice — reuse or clone; acronyms spelled as they must sound, locked in a dictionary.
- Script in parts — the page's argument, 6 to 10 pieces of 15 to 30 seconds.
- One paid master clip — about 1.20 EUR, once per person; the living eyes.
- The factory — repaints only the mouth per recording, under a centime a part. Drawn, step by step.
- Cut-out in pairs — one file for Chrome, one for Safari, always both.
- Page wiring — idle breath, card, icons, the slow glide; then the verification walk.
What it costs
| Piece | When | Cost |
|---|---|---|
| Master clip | Once per person | About 1.20 EUR |
| Each spoken part | Per segment | Under a centime (rented GPU) |
| Voice recording | Per script | ElevenLabs minutes, or free with Voicebox |
Where the material lives
| What | File |
|---|---|
| The whole recipe (the playbook) | site-speaker skill |
| Picture → talking video → cutout | photo-avatar-video skill |
| The talking feature, when ready | voice-agent skill · Voice |
| Frank himself explaining a document | frank-avatar skill |
| The recipes and the rules | Make pages move · criteria |
| The factory | the factory, drawn · musetalk.md |
| Valentina's page, clips, recordings | valentina-video.html · busto/ · narracion/ |