# Speech (/docs/localize/speech)





Narration turns text into audio, so learners can listen to the book — essential for learners with visual impairments or reading difficulties, and valuable for anyone who learns better by listening.

ADT Studio generates **localized text-to-speech**: natural narration in the right voice for each language of your ADT.

## Decisions to make [#decisions-to-make]

* **Which content is narrated?** A **Read-Aloud Content** setting lets you include or exclude each category independently — page text, image captions, activity answers, glossary entries, and Easy Read text.
* **Does the voice fit the audience and language?** A voice that sounds natural in one language may not be available or appropriate in another. Listen to samples before generating the full book.
* **Check the pronunciation of names and local terms.** Text-to-speech can stumble on proper names, place names, and curriculum-specific vocabulary. Spot-check sections that contain them — or replace just that entry's audio (see below).
* **Do readers need word-level highlighting?** Turning it on calculates word timestamps automatically during generation, so the reader can follow along word-by-word as it's narrated.

## Configure and generate [#configure-and-generate]

Speech needs your translated content ready first. Open the **Speech** stage and set a voice per language and provider, plus:

* **Word-level highlighting** — when enabled, word timestamps are calculated automatically as speech is generated. If you leave it off, you can still calculate timestamps manually later from the Speech view.
* **Read-Aloud Content** — toggle which categories of text get narrated at all.

<img alt="The Speech stage before it has run, showing Provider and Reader Highlighting settings alongside a sample audio preview" src="__img0" />

## Review, replace, and edit [#review-replace-and-edit]

Once narration is generated, browse entries per language, filter to just the ones still **Missing** audio, and preview each version with the version picker. If AI narration mispronounces a name or term, you're not limited to regenerating — every entry has an **Upload your own audio file** action (or **Replace this audio file** if one already exists), so you can substitute a manually recorded clip for that one piece of text without touching anything else.