A messenger assumes you can read. For a good part of a family — the youngest, and anyone who finds small text hard — that assumption quietly shuts the door. So Shoal can now read a message aloud. Tap the speaker, and it speaks. And for children using the Reef interface, text messages simply arrive as bubbles they can listen to, the way a voice note does — no reading required.
On the device, naturally
This is the companion to the choice we made with voice-note transcription, and it lands in the same place: the work happens on your device, never ours. The text is already decrypted on the device that’s about to speak it, so the app hands that text to the speech engine the operating system already provides and plays the result. Nothing is uploaded, nothing is synthesised on a server, and because there’s no round trip, it works offline. We don’t store an audio file for it either — the speech is made fresh, locally, each time, and then it’s gone.
A friendlier voice, if you want one
The built-in system voices are fine, and on some platforms very good — but they vary, and on a couple they’re a bit robotic for a small child to enjoy. So there’s an optional, warmer neural voice you can switch on. It mirrors how our on-device transcription already works: a one-time model download (gated by Data Saver so it never sneaks onto a metered connection), and after that it runs entirely on the device, using the graphics hardware where available and falling back gracefully where it isn’t. It’s off by default, and turning it on changes nothing about the privacy of it — the friendly voice reads the same already-decrypted text, in the same place, with nothing leaving the device.
There’s now a single picker for all of it: your device’s own voices and the friendly ones in one list, with a preview button so you can hear a candidate before you commit a child to it. We also warm the friendly voice up in advance, after an early version’s first word took an embarrassing few seconds to arrive.
Some honest limits
System voice quality is the platform’s, not ours, so it differs from device to device. The friendly voice asks for that one-time download before it’ll work, and it’s heavier on older hardware. And speech is an aid to reading, not a translator — it speaks the language the text is in, in the voice you’ve chosen, and doesn’t try to be cleverer than that.
How this fits with everything else
Read aloud is what makes Reef work for a pre-reader, and it sits alongside on-device voice-note transcription as the two halves of the same idea: turning speech into text, and text into speech, both without our servers in the loop. The security page describes what stays on the device, which here is all of it.
Try Shoal, hand it to someone who isn’t reading yet, and tell us how they get on at [email protected].