🤖 AI Avatar Reader — User Guide

1. What it does

On every article page there is a green Listen button near the top of the reading column. Press it and a small 3D anchor appears in the bottom-right corner and reads the article aloud, while the article itself stays fully visible behind and beside them, so you can read along, or just listen. There is one anchor — Divya, this publication’s own — and the site default and the choice saved in your profile decide which voice she reads with; see Which anchor reads to you below.

• Reads the title, then the summary, then the full body, in that order — exactly as an anchor would introduce a story.
• Reads everything the page shows, in the order it shows it — including the text inside picture-with-text blocks and each slide of a web story. Parts with no words of their own, such as an embedded map or post, are passed over quietly rather than described.
• No card, no panel, no border. The anchor floats over the page on a fully transparent background, so the text next to her is never hidden by a box.
• The text underneath stays selectable and scrollable — the overlay lets clicks and drags pass straight through to the article.
• She sits just above the chat button in the corner, so neither covers the other.
• Playback keeps going while you scroll.

While the 3D scene is loading you see a small spinner, not a face — the anchor never changes character mid-sentence. If your browser cannot run 3D at all, see section 5: the article is still read aloud, with no avatar.

2. The controls

Every control shows a short explanation when you hover or focus it.

Which anchor reads to you

This publication ships one anchor. The choice is still recorded per reader, because it decides which voice reads with her:

If you are signed in: open Profile → Preferences and use the Reading Avatar setting. It carries its own explanation on hover. Your choice is saved to your account, so it follows you to any device you sign in on, and it applies the next time an article page loads.

If you are not signed in (or have never made a choice): you get the site-wide default, which is chosen for the publication as a whole.

Why only one? A second, male anchor was built and then withdrawn, because it could not be animated correctly — it appeared from the wrong angle instead of showing a face. An option that cannot be shown properly is worse than not offering it, so only the anchor that displays correctly is offered. Your setting and the site-wide default both remain in place, so a second anchor can be added later without any rework.

3. Reading the status

The small line under the status appears only when it has something to add — a hold, a failure, or a mouth that cannot follow the voice. See the next section.

4. How the voice is produced

The anchor speaks in a natural human voice, in the language you are browsing in. There is nothing to choose or set up — press Listen and she starts reading.

Producing the voice takes a moment the first time a given article is played. After that the audio is kept, so replaying it — or another reader listening to the same story — starts immediately.

5. How the lip-sync works — and its limits

The anchor is a real 3D head rendered live in your browser, not a video and not a picture. When audio comes the browser analyses the live sound and moves her mouth from the result. Nothing is pre-rendered, so any language works — including Gujarati.

Her mouth opens and closes with the sound, and her teeth and tongue move with it. At the size she is shown, that reads as natural speech.

Being straight about what this is. The mouth is driven by loudness and frequency bands — that is, it moves the way speech moves, but it does not know which Gujarati letter is being spoken. It is convincing motion, not frame-accurate lip-sync.

If only the browser’s built-in speech is available, its audio cannot be analysed, so her mouth cannot follow it. The line under her then says “Estimated mouth movement” rather than claiming a sync that isn’t happening.

If 3D cannot run on your device

The 3D view needs 3D graphics support in your browser. Where that is switched off, blocked, or the model cannot be downloaded, the reader does not substitute a different-looking avatar. It plays the article as audio only and shows a small speaker mark in place of the anchor, with the line “Playing audio only”. Play, pause, stop and close all keep working.

Her appearance

She is dressed as this publication’s own anchor rather than as a stock 3D figure, and she is shown head and chest so her expression reads: long dark hair falling past her shoulders, a deep-crimson bindi on her forehead, a gold chain at her collar, and a hot-pink silk saree with a gold border.

Every colour is measured from the publication’s own reference photograph, not chosen by eye. Three of them are worth knowing:

What is applied is a brightness match, not a repaint: the model’s own texture is scaled so its average lands on the reference’s, which leaves every individual detail — pores, folds, the colour of the lips — intact. Repainting each pixel from its brightness alone is what makes a face look flat, and is exactly what the previous version did.

The earrings are real geometry, not a painted detail: a gold stud on each lobe, a flared bell hanging from it and a bead below that — the bell’s outline is what makes it read as a jhumka rather than a dot. Their position is measured from the head itself, at the widest row of the head mesh, which is where the ears are; a guess would have left them floating beside her.

It is not only colouring: her facial structure is reshaped as well. The stock model behind her is a generic northern-European head, so the geometry itself is adjusted to a South Asian structure — broader cheekbones, a fuller jaw that no longer tapers to a point, a softer chin and a wider nasal base. The change is deliberately small (a few millimetres per side, capped at 6 mm) because a larger push on a low-polygon head reads as a broken model rather than a different face.

Two areas are held completely still while her face is reshaped, because they would otherwise break: the eye sockets, so the eyeballs, lashes and brows keep lining up with the skin around them, and the lips, which sit directly over the teeth. Both are sized to the geometry they protect — the socket to the eyeball’s own reach, the lips to the mouth’s own opening.

Her bindi and chain are attached to her head and neck rather than fixed to the screen, so they stay in place when she turns her head while reading.

The shot is framed from her own measurements rather than from a fixed camera setting: when she loads, her height is measured, the camera is aimed from the top of her head and the bottom of the shot is set just below her chest. So there is always a little clearance above her head, she is never cropped at the top, and the frame is spent on her face rather than on the rest of her. The framing corrects itself if the avatar is ever replaced.

6. Languages

The avatar reads whichever language you are browsing in. Gujarati is the primary language, followed by Hindi, Marathi and English, all with the same controls. She speaks every one of these languages.

7. Privacy — what the reader does and does not do

• No account needed. The reader works whether or not you are signed in.
• Nothing you read is tracked. Which articles you listen to is not recorded against you.
• The lip-sync is calculated on your own device. The mouth movement is worked out in your browser from the audio as it plays, not sent anywhere.
• Turning text into speech sends the article text to a speech service and receives audio back. That is the only thing that leaves your device, and it is the text of the article only — nothing about you.
• Your reading position and your saved lists are yours. They are only stored when you are signed in, and only against your own account.

There is nothing to install and nothing to switch on — press Listen.

Troubleshooting

Related guides

• 📰 Articles — the article page the reader lives on.
• 📰 News listing — browsing and search.
• 📚 All user guides