Fish Audio Studio for Long-Form Narration

What long-form audio actually needs

A thirty second clip is a demo. A ten chapter audiobook is a production, and the problems are different: the voice has to stay the same from the first page to the last, the pacing has to survive hours of listening, and a single bad paragraph should never mean re-generating the whole file.

Long-form work in Fish Audio is built around those three constraints. The voice is chosen once and reused, the script is the source of truth rather than the audio, and generation happens in paragraphs or chapters so that fixes stay local. The faster queue and API access that come with the Studio pack exist to make that volume manageable rather than to change the output quality.

If you are still choosing a voice, start with the voice library. Committing to a narrator after a full paragraph, not after a sentence, is the single most common piece of advice for a reason.

Structuring a long script

Write first, generate second. A script that still changes while you generate is a script you will pay for twice.

Practical structure looks like this:

1. Split the work into chapters, and chapters into paragraphs of a few sentences.
2. Keep each paragraph a separate generation, with its own file name, so a re-take replaces one paragraph.
3. Mark delivery changes inline with emotion tags instead of describing them in a separate document.
4. Generate in reading order, and listen as you go: fixes are cheapest while the surrounding text is still fresh.
5. Keep the final script and the final audio in the same folder, so a revision months later is a text edit rather than a rebuild.

This is the same workflow whether the project is an audiobook, a set of explainer videos, or a course. The text to speech guide covers the single-line case if you only need short pieces.

Emotion tags and narration styles

Emotion tags are what keep a long read from flattening out. They are markers written into the script itself, so direction stays attached to the text: the pause before a reveal, the shift to a warmer register in a dialogue section, the drop in pace at the end of a chapter.

For narration specifically, restraint usually wins. A narrator carries a book by staying consistent; dramatic delivery in chapter one becomes exhausting by chapter ten. Character voices are the opposite: they are chosen for personality, and the tags are what let that personality change line by line within a scene. Both approaches run on the same engine as cloned voices, so a project can mix a trained narrator with a specific character without leaving the workspace.

Credit planning for a long project

Credits are predictable once you know the exchange rate: one credit per 1,000 characters. A paragraph of roughly 100 words is about 600 characters, so about 0.6 credits. A chapter of 1,500 words is roughly 9,000 characters, or about 9 credits, and a ten chapter book lands near 90 credits before any re-takes.

That makes re-takes the variable, not the writing. Budget for them by keeping paragraphs small and by proofreading the script before generating rather than after. The Studio pack carries 15,000 credits, which is far more than a single book needs; the smaller Starter and Creator packs are the better fit for a chapter at a time. Every pack is a one-time purchase, and credits stay valid for 12 months, so a long project can be spread across a quarter without anything expiring. The pricing page lists the packs, and the free tier guide explains what is available before you buy.

Frequently asked questions

Can Fish Audio generate an entire audiobook in one go?

It can generate the audio for one, chapter by chapter. Working in paragraphs keeps each generation fast, keeps costs visible, and makes a bad line cheap to replace.

How many credits does an audiobook need?

Roughly one credit per 1,000 characters generated. At 1,500 words per chapter, a ten chapter book is around 90 credits before re-takes.

Which pack is right for long-form work?

Creator covers most single books with 4,000 credits. Studio adds 15,000 credits, API access, and the fastest queue for teams producing continuously.