Character Voice

Character voice is what lets a reader assign a line to a speaker before reaching the name attached to it. It is not vocabulary and not a verbal tic. It is the register a person defaults to when the same pressure arrives: what they reach for, how much they concede, how quickly they get to the point.

The part that does the most work is negative. It is what this person will not say under that pressure, the subject they route around every time. Where interior narration is unavailable or the word list is short, this is the channel that still carries the inside of a person, because the reader hears what the character chooses to say.

Core takeaways

• A reader can assign a line to a speaker before reaching the name attached to it.
• The strongest marker is negative: what a character will not say, and the subject they route around every time.
• One character has to sound plain. Give every voice a strong tic and none of them registers.
• A character’s default register is how that character’s wound shows in speech.
• Under a ceiling of a few hundred shared characters, rare words and spelling accent are gone. Structure, silence, and topic-stance remain.
• The tag-strip test checks the whole claim. Remove the tags from a two-hander and see whether a fresh reader can assign the lines.

Ways one voice differs from another

Idiolect is one person’s total way of speaking: sound, words, syntax, discourse. The craft version is personal diction in word choice, rhythm, and word order, set so that no one else in the cast speaks that way. Sentence length, rhythm, refused topics, and default register under stress cost no additional vocabulary.

A listener sorts two leads by cadence alone: clipped flat clauses against long, exclamation-heavy runs.

Formality is another dial. The group addresses one adult in mock-military rank: salute, order, soldier. The gag also puts old-school vocabulary into a beginner’s book.

Relationship to certainty is a third dial. 阮草 asserts flat and does not qualify. mát tay names the hand that already knows what a living thing needs, and her bluntness is that trait in speech. Her voice is the cleanest in the cast, so it is heard most.

Signature form is a fourth dial. 小圖 speaks in four-syllable idiom-shaped bursts run through an old-telephone filter. The tag is never needed. The shape is the shape of a four-character idiom, not necessarily a real 成語. Do not restore dispensed-wisdom content to those bursts.

Silence and topic-avoidance cost no vocabulary and sit at the center of the low-level books. A voice lives in which of three it defaults to: the Said (words spoken), the Unsaid (conscious thought withheld on purpose), the Unsayable (the thing that cannot be voiced at all). 沈文 never states that his home was cold. He goes quiet whenever family comes up. He lets it slip and never announces it.

A published differentiation checklist names vocabulary, throwaway words, tight versus loose phrasing, and grammar: speech matching background, no two characters talking alike. Under a Han ceiling, what a character declines to say is the cheapest entry on that checklist.

In audio, the instruct, tired or flat or delighted, carries interiority the lexicon cannot buy. That is how a line is marked once voice is treated as heard and not only as written.

The plain center

The cast needs an unmarked center. 沈文 is built as the baseline, the everyman point of view, the neutral standard the others bend away from. That flatness of register is necessary. Give every voice a strong tic and none of them registers. Deviations need a norm. 白龍’s clipped deadpan, 星野’s exclamation runs, and 小圖’s four-syllable bursts all read as marked because a plain center exists.

Default register and the wound

A character’s default register is how that character’s wound shows in speech. 白龍’s deadpan covers a loneliness he will not admit, and the clipped voice is that loneliness in audible form. The wound is the interior cause. A crack in a fixed register reads as a reveal on its own, without new exposition.

When two registers meet, the clash of values shows in the speech itself. 星野’s over-enthusiasm meets 林薇’s dry local register. Each one romanticizes the place the other wants to leave. A two-hander spent on that pair characterizes both at once.

Flat characters and the one note

A flat character carries one memorable note, recognizable on every entrance. 蘇老師 asks the calm question everyone was avoiding. 吳老闆 is deadpan and avuncular. 將軍 is the rank gag. One-note voices leave the limited roundness budget for the leads who carry arcs. Under a tight word ceiling they are useful. Round Characters and the Telling Detail sets that roundness budget.

What survives a few hundred characters

Diction narrows to a few hundred shared characters. No one can be marked by rare words the cast has not learned. Structure and content stay free. The vocabulary ceiling is that early-book budget: a few hundred shared characters, actions and objects, almost no feeling-words. It is a house parameter and not a language-test law. Story Under a Vocabulary Ceiling explains why structure and silence are the only voice available early.

Spelling-based accent is the usual cheap differentiator. It has no existence in Han. The script is not an alphabet, and eye-dialect does not transfer.

Register can widen as the reader’s vocabulary grows. The same fixed form can carry heavier content. 小圖’s bursts stay broken in form while what they land on sharpens. The form stays fixed, and the bursts do not become oracle-lines. The widening also shows the reader that their own vocabulary has grown. Ensemble banter carries the high-frequency repetition that story-listening relies on while the voices stay distinct.

At two hundred and fifty characters, a beginner parsing the sentence has little attention left for nuance. Marking has to be coarse enough to notice on a first read. The speaker cap is at most three or four speaking voices in one reading, so a listener can tell who is talking.

Gesture and object and action are the other channel when the sentence has no room for a mark.

What it costs and where it fails

The benefit is that a reader can assign a line without the tag. The cost is tracking one habitual register per character, a differentiation gate at authoring, and which book has widened whose register. No purchased sheet is required. The tag-strip test and the dials above replace it.

A tic hardens into shtick if it is not watched. The enthusiastic character becomes nothing but enthusiasm. Voice carries little early on. Using it to carry interiority the words cannot name produces mannerism and no depth. At two hundred and fifty Han, with three short readings and a four-speaker cap, marking every voice heavily leaves a cast of tics. Characters who should stay flat, 將軍 and the adults, gain nothing from a full idiolect.

Where a scene turns on meaning the reader is still decoding, a distinct voice is a luxury. Carry the beat by object and action.

Tests and quit signals

Strip the tags from a two-hander. If a fresh reader cannot assign the lines, fix content and topic-stance before adding more tics. If the reader names the speaker but winces, dial back toward the baseline. If a scene only works once the withheld thing is said outright, return it to the Unsaid and let a curious friend do the circling. A curious friend is the second person who circles a withheld thing until it can be said.

The claim is checkable. The tag-strip test is the falsifier. A deadpan that cracks should read as new. The cold house is never stated. Two designed voices in audio should be told apart by cadence before content.

A reader can still assign lines after the ceiling has taken spelling, rare words, and the tic-for-everyone option. What is left is the default register the wound produces.

How to practice this

1. Strip the speaker tags from one two-hander and hand it to a fresh reader. Notice whether they can assign every line.
2. When the reader cannot assign the lines, change content and topic-stance first. Notice whether your first instinct was to add a tic.
3. Pick one character and list the subjects they route around under pressure. Notice which subject never appears in their lines.
4. Check that one character in the cast has a plain register. Notice whether the marked voices still read as marked.
5. Count the speaking voices in one reading. Notice whether more than four people are talking.
6. Play two designed voices as audio and ignore the words. Notice whether cadence alone separates them.

Related pages

• The Wound and the Lie: the interior cause. The default register is how the wound shows in speech.
• Interiority Through Action and Object: the other channel when the sentence has no room for a mark.
• Story Under a Vocabulary Ceiling: why structure and silence are the only early voice.
• Round Characters and the Telling Detail: roundness and the single fixed note, the budget voice inherits.

Open questions

• Whether the audio instruct should be marked on the line once voice is treated as heard, not only as written.

Sources

• Robert McKee, Dialogue: The Art of Verbal Action for Page, Stage, and Screen (2016): personal diction; Said / Unsaid / Unsayable.
• Sol Stein, Stein on Writing: differentiation checklist: vocabulary, throwaway words, tight versus loose phrasing, grammar.
• E. M. Forster, Aspects of the Novel (1927): a flat character carries one memorable note.
• Han as a non-alphabetic script: spelling-accent has no existence in it. A script fact, not a citation.

The unmarked-center law, the wound-print, and the ceiling parameters (two hundred and fifty Han; four-speaker cap) are this vault’s teaching defaults (2026-07-09 / 2026-07-20).