Narration from the translated text
Speech generation voices a licensed translation chapter by chapter, in one professional voice per language held for the whole book.
For audiobook publishers
Most translated editions never get an audiobook, because narration in a smaller market rarely recovers its studio cost. Octavia produces narration from a licensed translation, or carries existing narration into another language with the narrator’s consent — chapter by chapter, with every line reviewable before it is voiced.
The specific reasons this content usually stays in one language.
A studio, narrator and director cost much the same for a small edition as for a bestseller, so smaller-language editions stay silent.
Narrator and author agreements were written for a recorded performance, and most say nothing about a voice model built from it.
A name read two ways, a chapter brighter than the last — over nine hours, small inconsistencies become the reviews.
The parts of the platform that matter for this work, rather than the full feature list.
Speech generation voices a licensed translation chapter by chapter, in one professional voice per language held for the whole book.
With a narrator’s written consent, their own voice carries into the translated edition, with control over what may be produced in it.
Custom vocabulary keeps character, place and invented names from being translated or respelt, however often they occur.
Transcribe the finished narration and check it against the text, so dropped lines and misreads are found by search, not by re-listening.
Each chapter’s translation is editable and approvable before audio is generated — where an editor’s change is a text edit, not a retake.
Diarization keeps each performer distinct in full-cast titles; Magic Mode lifts narration off music and sound design.
Four steps from the file you already have to every language you need.
Translation rights, audio rights and any voice consent confirmed per title and language.
Work chapter by chapter, with house rules for headings, numbers and footnotes set up front.
Pin recurring names; a native-speaking editor approves each chapter before it is voiced.
Listen closely to the opening chapters, proof the rest against the transcript, then assemble per retailer.
One upload in, every format you need to publish out.
Including the ones where the honest answer is a limitation.
Probably not, so read it before assuming. A licence to record and distribute a performance is not a licence to build a voice model from it. Get written consent naming the titles, languages and term, what happens if consent is withdrawn, and whether the model may be reused — and expect narrators and their agents to ask first how listeners will be told.
Policies differ between retailers and are still changing, and some require synthetic narration to be disclosed in metadata or credits. Check the current terms of every store you distribute through before production rather than at upload. Whatever the rules say, state it in the credits: listeners accept disclosure far more readily than discovering it afterwards.
It can translate narration, but the right to translate the text is a separate licence from the audio rights, and you need it either way. For fiction, have the translation written or line-edited by a literary translator — a flat rendering is far more audible over nine hours than on the page. Where a licensed translation already exists, voice that rather than translating the audio again.
Proof the text wherever you can. Approve the translation chapter by chapter, search the script for every pinned name to confirm it appears consistently, then compare a transcript of the finished audio with the text to catch dropped or misread lines. Save close listening for the opening chapters, dialogue-heavy scenes and the first appearance of each name.
Practical guides from the Octavia editorial team.
The same engine, pointed at a different kind of work.
Start with one backlist title that already has a licensed translation and a clear rights position, and proof it chapter by chapter.