Tag every line of an argument [angry] and the scene comes back hot from the first word to the last. Nobody gets to change their mind, because the tag never switches off. Take the tags out, put one [emphasis] on the word that actually matters, and the same scene sounds like two people arguing again. Most good direction works this way: less of it, placed more exactly.

Directing a text-to-speech voice is mostly restraint. Punctuation does the daily work. A tag is for the moment punctuation cannot say. VoxAnvil does not give you a booth talkback. It gives you the manuscript, a set of tags, and a synthesis step that uses a third-party voice provider. You are not booking a session with that provider, and nothing here is a joint product. What you are doing is preparing text.

Start with punctuation

A period is a stop. A comma is a small hinge. A question mark lifts. If the read is flat, look at the punctuation before you look at a tag. Join two short sentences and the voice may rush. Split one long sentence and the voice may recover.

Use a paragraph break when the scene or the topic actually changes. A paragraph break is a bigger pause than a comma, and it gives a chunk boundary a clean place to land later. Do not insert blank lines just to "add emotion." You will hear a hole.

Quotation marks tell a voice that someone is speaking, but they do not invent a second actor. One narrator still reads both parts. If you need the listener to know who spoke, keep the dialogue tag in the sentence. "She said" is not dead weight in audio. It is a handrail. More on that is in the dialogue piece.

Pause tags, and what happens to them

The studio toolbar includes pause marks such as [beat], [pause:0.5s], [pause:1s], [pause:3s], [pause:5s], and [long-pause]. Before the text is sent for speech, those tags are converted. [beat] and pauses under 2 seconds, including [pause:1s], become an ellipsis. Pauses of 2 seconds or more, and [long-pause], become an ellipsis followed by a paragraph break. The tag is not left in the sentence for the voice to read as the word "pause," and it is not a sample-accurate silence you can measure on a timeline.

That limit matters. If you need a timed gap for a video edit, leave the gap in the editor. Do not expect [pause:3s] to be three silent seconds of mastered room tone. Expect a paragraph break in the narration. For an audiobook-style chapter, a short pause at a turn is enough. A five-second hole in the middle of a page sounds like a dropout.

Use one pause where a human would actually stop: before a reveal, after a list item you want remembered, between a question and the answer in a lesson. If you can count more than a few pause tags on a page, you are directing the page instead of writing it.

Expression tags, used once

These expression and performance tags are recognized in the studio and kept in the text that goes to synthesis: [angry], [sad], [excited], [emphasis], [whispering], [soft], [breathy], [embarrassed], [chuckle], [laughing], [sighing], [crying loudly], [sobbing], and [groaning]. The speech engine is expected to interpret them. Results vary by voice. A whisper tag on a bright voice may only get quieter. Listen to the line. If the tag did nothing you can hear, take it out rather than stacking a second one.

Put the tag at the start of the short span you mean, not at the start of the chapter. A tag that covers a page locks the performance. A tag on a single sentence lets the next sentence return to the narrator's normal read, which is what you want in almost every scene.

[emphasis] on one word beats [excited] on a paragraph. Emphasis is a pointer. Excited is a mood, and mood tags get old.

Unrecognized tags are a liability. Narration QA on the Pro plan warns on a square-bracket tag it does not know, because the engine may speak the bracket or ignore your intention. If you type [whisper] instead of [whispering], fix the word. Do not invent a tag and hope.

What not to direct

Sound-effect tags such as [sfx:rain:loop] can sit in a manuscript, and a well-formed one is treated as a known tag by the narration check. The current speech path removes those tags before synthesis. They are not spoken, and they are not mixed under the voice. If you want rain, add rain in an audio editor after you export. Writing the tag will not build a sound bed. This is the honest limit of direction inside the chapter text.

Do not direct by shouting. All caps is a blunt instrument. Do not direct by repeating the line three times unless the book repeats it on purpose. The narration check flags a repeated sentence because accidental duplicates are common when you paste a revision under the original and forget to delete one.

A light hand on one scene

Take a scene in which a character finds an empty driveway. The plain sentences already carry it: "The driveway was empty. She checked the street. Then she checked it again." If the second check needs air, add [pause:0.5s] before "Then," generate that paragraph only, and compare it with the untagged read. Keep the take in which the pause sounds like a person looking, not like a file that glitched. Leave the rest of the page alone.

When a whole chapter feels monotone, the fix is usually sentence length and a better voice match, not a tag on every line. Audition with clean text first, using the matching steps. Add direction only where a specific line failed.

Frequently Asked Questions

Q: Can I stack several expression tags on one line? A: You can type them. You should not. One tag on a short span is something you can judge by ear. Three tags is a guess you cannot debug.

Q: Will a pause tag produce a measured silence? A: Not as a studio timeline. Pause tags under 2 seconds become an ellipsis. Pause tags of 2 seconds or more, and [long-pause], become an ellipsis followed by a paragraph break. Measure a video gap in your editor.

Q: Do these tags work in every speech product? A: No. This guide describes the tags VoxAnvil recognizes and how this studio prepares them for synthesis. Another tool may speak the brackets or ignore them. Read that tool's own notes before you reuse a tagged script.

Q: Should every audiobook chapter be tagged? A: No. A clean sentence in a steady voice is the default. Tag the line that needs a change you can name.

On your next chapter, strip every tag, listen once, and put back only the ones you can justify in a sentence. Usually that leaves one or two.