On the page, a line break tells you the speaker changed. In audio, a line break is a short pause and nothing more. Four quick exchanges into a kitchen argument where both characters are angry and every line is short, a listener can lose track of who wants to leave. The voice itself is fine. The text never said the names, because on paper it didn't have to.
Dialogue flattens when the listener has to do the bookkeeping the page used to do with indents. A single narrator is the normal path in VoxAnvil. Multi-voice casting is optional, and it is on the Pro plan. You can get a clear scene without it if the sentences carry the speakers.
Put the speaker where the ear can catch it
On the page, you can write a name once and then run five naked lines. In audio, reintroduce the speaker when a listener could lose the thread. That is often every two or three exchanges, and always after a block of action.
"Said" is a clean word. It disappears, which is what you want. Fancy verbs such as "snarled" and "opined" stick out and ask the voice to act. If you want the line to sound angry, the words inside the quote should be angry. The tag can stay plain.
Place the name where it saves the listener, not always at the end. A leading tag helps when the line is short: "Mara said, 'Take the keys.'" A trailing tag helps when the line is a sentence you want heard first. Do not bury the name in a long beat. The listener is waiting to know who is talking.
Quotes, beats, and action
Keep real quotation marks around spoken words. A dash can work in print as a dialogue marker and then disappear into a pause. If your manuscript uses dashes instead of quotes, listen to one page before you commit. If two speakers blur, convert that page to quotes and names and compare.
Action beats are your second tool. "She put the keys on the table" between two lines gives the ear a rest and a camera. One beat is plenty. Three beats in a row, each with its own adverb, and the scene turns into stage directions. The voice will read them, because they are sentences.
Do not leave a pronoun to do a name's job when two people share the pronoun. "She told her" is a puzzle. Names are cheaper than confusion.
Interruptions need a decision. An em dash at the end of a cut-off line often reads as an unfinished sentence, which is what you want. An ellipsis reads as a trail-off. Pick one and use it for that meaning only. A page of mixed dashes and dots sounds drunk. If you want a short silence before the reply, a [pause:0.5s] at the start of the reply is clearer than another ellipsis. Remember that the studio turns that pause tag into a short break in the text, not a frame-accurate gap.
One voice first, a second voice only when it earns the seat
Generate the scene with one narrator before you cast anyone else. If the scene is clear with names and beats, stop. A second voice is another performance to keep consistent, another pronunciation list, and another thing that can drift in chapter twelve.
When you do cast, use character casting on Pro. The studio can read an excerpt, propose speaking characters, and store a cast with a voice assignment. Assign voices that are easy to tell apart in the same headphones. Two calm midrange voices will collapse back into one. Give the second voice an obviously different pace or pitch, then read a page where they share a short exchange. If you still need the names, keep the names. Voices do not retire the dialogue tags. Listeners tune out, and the tags bring them back.
Do not give every minor character a voice. The barista with one line can stay in the narrator's mouth. Casting is for people the listener must track.
A page you can hear
Consider an exchange written for print: "You left." "I came back." "Late." On the page, the indents do the work. For audio, write: "You left," Mara said. Jon said, "I came back." Mara set the keys down. "Late." The last line can stay bare because the action named her a breath earlier. Generate that version and the bare version with the same voice. Keep the one you can follow from the hallway, which is the honest test. If you cannot follow it while you walk away from the speakers, a tired listener will not follow it either.
Expression tags are optional and dangerous in dialogue. One [angry] on a single short line can work. The same tag on both speakers makes them the same person at a higher volume. Direct the words first. The pause and emphasis guide covers how little tag you actually need.
Also watch repeated lines. Writers duplicate a speech when they revise and leave the old one underneath. Narration QA on Pro flags a repeated sentence. Even without the report, search the scene for a line that appears twice before you generate.
Frequently Asked Questions
Q: Should I remove all dialogue tags for a more cinematic read? A: No. Cinema has faces. Audio has names. Remove a tag only when the speaker was just named and the next line is short.
Q: Can italics show a thought instead of speech? A: Not reliably. If the listener must know it is a thought, write "she thought" or fold the thought into narration. Do not depend on italics to change the voice.
Q: Is a full cast better for sales? A: This guide will not pretend to know your sales. A full cast is more to direct and easier to get wrong. Add voices when the scene stays confusing after the punctuation pass.
Q: What if my characters share a dialect? A: Do not spell a dialect in a way you would not want read letter by letter. A few word choices carry more than phonetic spelling, which a speech engine may over-pronounce. Keep the grammar you intend to be heard, and test one paragraph.
Cover the speaker names on the page and listen to the scene. Wherever you lose track, add a tag, an action or a name, and leave everything else alone.