Synthetic voices

Also called TTS voices, Computer voices, Neural voices, System voices

Where computer voices come from, why the good ones are already on your phone for free, and how to choose one that sounds like you.

What it is

A synthetic voice is the particular voice a text-to-speech engine speaks in: its sex, age, accent and character. On a phone or tablet they come from the operating system; on a dedicated device they come from the manufacturer's licensed voice library. Some are free, some are bought, and a few are made from a real person's own recordings (voice banking).

A voice is not a setting. For a person who will use it for years, it is part of how the world hears them, and it deserves the same care as choosing a name.

Who it can help

Everyone whose aid speaks. It matters most for people who will use one voice for a long time and in public: children, whose voice should be a child's and should grow with them; adults with a strong regional identity; people whose accent, sex or age is not the default the app shipped with.

It is a poorer answer for people who want their own voice back, for whom voice banking or message banking is the real thing. And the most natural voice is not always the most intelligible one on a noisy bus; some people keep a clearer, plainer voice for out and about.

How it works

Modern voices are built from many hours of one speaker's recordings and a model that learns to generate speech in that voice from any text. On Apple devices, the higher-quality British English voices are downloaded through the accessibility settings; on Android they come through the speech engine's own settings; on Windows through the language packs. All free, all usable by any app on the device.

A voice has a few properties the person controls: speed and pitch, sometimes volume separately from the device volume, and on some systems a pronunciation dictionary. Children's voices exist but are scarce, and a young child is often better served by a light, higher-pitched adult voice than by a caricature. Regional British accents are increasingly available; ask for them.

Getting started

Sit with the person and a list of five sentences they say often. Play each one in four or five voices, and let them react. Don't choose for them, even for a young child; watch the face. Note which they lean towards and which they wince at.

Then download the higher-quality version of the chosen voice (it will sound noticeably better than the compact one), set it in the app, and write the voice name down in the communication passport so it survives a device reset or a new tablet.

Where to get help

AbilityNet walks through the voice settings on every platform. Your AAC service will help match a voice at an AAC assessment, and Communication Matters is where AAC users talk about the voices they actually use. For a voice made from your own, see voice banking and the MND Association.