Text-to-speech

Also called TTS, Speech synthesis, Speech output

How a device turns the words on the screen into a voice in the room, and what to check when the voice says the wrong thing.

What it is

Text-to-speech is the software that reads text aloud. A communication aid takes the words a person has chosen, whether by symbol or by spelling, and hands them to a speech engine, which works out how to say them and plays the sound. It is the difference between a board people have to read and a voice people can hear across a room.

Every speech-generating device has one. So does every phone and tablet, built in and free, which is why a communication app can speak without bringing its own.

Who it can help

Anyone who uses a text- or symbol-based aid and wants to be heard rather than read: people who are non-speaking, people with dysarthria whose speech is hard to understand, people after laryngectomy, people with aphasia who can select but not say.

It is not for everyone. Some people prefer their own recorded voice (message banking), some Deaf users of text prefer the screen, and some people find any synthetic voice hard to accept as theirs; that is a real objection and choosing a voice with care is the answer to it.

How it works

The engine first normalises the text (turning "3pm" into "three p m", expanding "Dr"), then predicts pronunciation from spelling and a dictionary, then generates the sound. Older engines stitch recorded fragments; newer ones generate the waveform from a model, which is why modern voices sound so much more natural.

The voices come from the operating system on a phone or tablet (synthetic voices explains where from) and the app chooses among them, then sets speed and pitch. Two practical facts: pronunciation of names and places is fixed with a pronunciation dictionary on most devices, and the best voices on a phone are often downloads that are not installed by default.

Getting started

Open the accessibility or spoken content settings on the person's phone or tablet, look at the voices installed, and download the higher-quality British English ones; they are free and they sound like people. Then pick one in the communication app and let the person hear three or four and choose.

Type the person's name, their street and the family dog into the app and speak them. If any is said wrongly, find the pronunciation setting and fix it now; being introduced by a voice that mangles your name is a bad start.

Where to get help

Your AAC service will set up voices as part of an AAC assessment. AbilityNet explains the spoken-content settings on every platform, and Communication Matters publishes plain-language guides on speech output. The RCSLT is the professional body for the therapists who advise on it.