In nonfiction narration, the riskiest small word is "not." A voice that would be lovely in a novel can lean on the wrong word in a paragraph about a budget line and make a neutral sentence sound sarcastic. A policy report has no use for sarcasm. It needs a reader who can say "not the 2019 figure" in a way the listener trusts was intentional.
Nonfiction narration is a clarity job. Warmth is always welcome. Performance is a risk. When you audition voices for a reported book, a memoir of work, a textbook, or a business guide, test the sentences that would mislead a listener if they were stressed wrong. A demo reel of emotional range will not show you that.
The four sentences that matter
Build the audition from four sentences you already wrote. Do not write a special "voice test" in a style the book does not use.
- A definition. One sentence that says what a term means. The voice should sound like it is explaining, not unveiling a secret.
- A number sentence. Include a year, a quantity, and a unit you actually use. "In 2019 the agency counted 14,000 cases" is a better test than "things got worse."
- A name sentence. A person, a place, and an institution. If the voice smears the institution, it will smear the bibliography too.
- A careful sentence. Something with "not," "only," or "less than" in it. These words are small and load-bearing. If the voice throws them away, the book changes meaning.
Generate all four with the same voice and no expression tags. Tags add attitude you are not auditioning yet. You are auditioning whether the voice can carry a plain sentence.
What to score
Score each sentence pass or fail before you score the voice as a whole.
- Stress. In the careful sentence, is the negation audible? Play it once and write down what you think the sentence claimed. If you write the opposite, the voice fails, even if it is pretty.
- Digits and names. Did the year sound like a year? Did the name survive? A failure here is a script problem if every voice fails, and a voice problem if only one voice fails.
- Authority without theater. Does the voice sound like a person who has read the material, or like an advertisement for the material? Advertisement voices make nonfiction feel unsourced.
- Endurance. Nonfiction chapters are often longer than stories and less varied. Play the four sentences twice. Irritation on the second play is a real score.
A voice can be warm and still fail stress. Keep the fail. You can find warmth in the next candidate. You cannot edit a missing "not" after export without regenerating the line.
How nonfiction scripts should be prepared first
Audition a script you would actually publish. Expand the abbreviations you would say out loud. "vs." becomes "versus" if that is the spoken form. Citations that are only "Smith 2019" need a decision: will the listener hear the citation, or will it live in a PDF notes file? If it is in the spoken file, write "Smith, 2019" as words you want said, or cut it from the audio edition on purpose. Do not leave a parenthesis full of initials and expect a graceful skip.
Tables do not narrate. If the book has a table, write two or three sentences that state the comparison you want heard. A voice reading column headers will sound like a spreadsheet, because it is reading a spreadsheet.
The studio preflight scan flags numbers, abbreviations, names, punctuation, currency, and dates, and it suggests a spoken replacement. Use that on the audition paragraph so you are not judging a voice for a symbol you meant to replace. Pronunciation Forge, on Starter and Pro, stores the reading you keep. The numbers guide walks through the decisions. Do that pass before the final audition, not after you have fallen in love with a take.
A fair shortlist
Run three voices on one sentence: "The program did not replace the 2014 benefit. It added a smaller one." Voice A sounds intimate and skips past "not" so the first clause sounds like praise. Voice B is plainer and both clauses stay distinct. Voice C is vivid and adds a lift at the end that sounds like a sales close. B is the narrator. A and C can still be useful for a trailer you write separately, with copy that is allowed to persuade. The book is not the trailer.
Limit the shortlist to voices you can keep for the whole book. Switching voices at chapter ten because a later chapter is "more exciting" breaks trust. Nonfiction listeners use the voice as a guide. Guides do not change faces at the hard chapter. If a quoted interview needs a second speaker, that is a different decision, and character casting on Pro can assign one. The narrator of the author's prose should stay put. See keeping a voice across a series if the book is one of several.
A third-party voice provider does the synthesis inside VoxAnvil. You are choosing among voices available to your plan, including the 10 curated narrators on the free plan. You are not getting a house style from a partner program. If none of the curated voices can say your careful sentence, fix the sentence and try again before you assume you need a clone.
Frequently Asked Questions
Q: Should nonfiction use a whisper or a soft tag for sensitive chapters?
A: Almost never as a default. A sensitive chapter still needs to be understood. Use plain delivery. If one sentence must be quieter, try [soft] on that sentence only and compare it with the untagged line.
Q: How long should the nonfiction audition be? A: Those four sentences are enough to cut the list. Then generate one full section of about five minutes with the finalist. If you are bored or lost in that section, do not generate the book.
Q: Can I audition with the acknowledgments? A: Acknowledgments are full of names and thin on argument. Use them as a name test, not as the only test. The careful sentence has to come from the argument of the book.
Q: What if the voice says a citation in a clumsy way? A: Decide whether the citation belongs in the audio. If it does, rewrite it as speech: "in a 2019 paper by Smith." If it does not, remove it from the spoken file and keep it in the ebook notes.
The test for a nonfiction voice is simple: play your most careful sentence once, then repeat the claim from memory. If you can, the voice is doing its job.