Performs a systematic analysis of every character's dialogue to measure distinctiveness and
identify where characters sound interchangeable. Produces quantified voice fingerprints and
actionable recommendations for differentiation.

## Key Points

- User says "my characters all sound the same" or "check my dialogue"
- User wants a voice audit, dialogue review, or speech pattern analysis
- As preparation for dialogue revision or a read-aloud pass
- When the Novel Audit's Module 5 flags voice consistency issues and deeper analysis is needed
- **The manuscript** — full text with dialogue clearly formatted (quotation marks)
- **Character bible** (optional) — helps attribute ambiguous dialogue lines
1. Extract every line of dialogue from the manuscript.
2. Attribute each line to a speaking character using dialogue tags and context.
3. Group all dialogue by character.
4. Flag lines that cannot be confidently attributed.
5. Calculate total dialogue word count per character.
- Unique word count / total word count (vocabulary richness ratio)

Dialogue Voice Auditor Skill

Performs a systematic analysis of every character's dialogue to measure distinctiveness and identify where characters sound interchangeable. Produces quantified voice fingerprints and actionable recommendations for differentiation.

When to Use This Skill

• User says "my characters all sound the same" or "check my dialogue"
• User wants a voice audit, dialogue review, or speech pattern analysis
• As preparation for dialogue revision or a read-aloud pass
• When the Novel Audit's Module 5 flags voice consistency issues and deeper analysis is needed

Input Requirements

• The manuscript — full text with dialogue clearly formatted (quotation marks)
• Character bible (optional) — helps attribute ambiguous dialogue lines

If dialogue tags are inconsistent or missing, the tool will do its best to attribute lines from context but will flag uncertain attributions.

Analysis Framework

Phase 1 — Dialogue Extraction

1. Extract every line of dialogue from the manuscript.
2. Attribute each line to a speaking character using dialogue tags and context.
3. Group all dialogue by character.
4. Flag lines that cannot be confidently attributed.
5. Calculate total dialogue word count per character.

Phase 2 — Voice Fingerprinting

For each character with 10+ lines of dialogue, build a voice fingerprint:

Vocabulary Metrics

• Unique word count / total word count (vocabulary richness ratio)
• Average word length, jargon usage, profanity frequency
• Contractions frequency (formal vs. casual indicator)

Sentence Structure

• Average sentence length and variance (monotone vs. dynamic)
• Question, exclamation, and fragment frequency
• Compound sentence frequency (rambling vs. concise)

Speech Patterns

• Verbal tics and filler words, catchphrases, greetings, farewells
• How they express agreement, disagreement, and deflection

Emotional Register

• Default emotional temperature (warm, cold, neutral, volatile)
• Range of emotions expressed in dialogue vs. expressed through action
• How they handle conflict in conversation (direct, passive-aggressive, avoidant, escalating)
• Humor style (sarcastic, dry, physical, self-deprecating, none)

Relationship-Dependent Voice

• Does the character speak differently to authority figures vs. peers vs. subordinates?
• Do they code-switch between settings?
• Are these shifts consistent throughout the manuscript?

Phase 3 — Similarity Analysis

Compare every character pair using the fingerprint dimensions:

1. Compute a similarity score (0-100) for each character pair.
   • 0 = completely distinct voices
   • 100 = indistinguishable
2. Flag any pair scoring above 70 as "dangerously similar."
3. Flag any pair scoring above 85 as "effectively identical."
4. Identify the specific dimensions driving high similarity (is it vocabulary? sentence length? emotional register? all of the above?).

Phase 4 — Differentiation Opportunities

For each high-similarity pair, recommend specific changes:

• Which character should change (usually the less-developed one)
• Which dimensions to adjust (the easiest levers to pull)
• Concrete examples: "Give Marcus shorter sentences and more fragments. Replace his current 'I think we should consider...' patterns with 'No. Bad idea. Next.'"

Output Format

# Dialogue Voice Audit Report
**Title**: [Novel title]
**Date**: [Today]
**Characters analyzed**: [N]
**Total dialogue lines**: [N]

## Voice Fingerprint Summary

| Character | Avg Sentence Len | Vocab Richness | Contractions | Questions | Fragments | Emotional Temp |
|-----------|-----------------|----------------|--------------|-----------|-----------|----------------|
| ... | ... | ... | ... | ... | ... | ... |

## Detailed Fingerprints

### [Character Name]
- **Total dialogue**: [N] lines, [N] words
- **Vocabulary**: [richness ratio], [complexity level]
- **Structure**: [avg length], [variance description]
- **Tics**: [list]
- **Catchphrases**: [list with frequency]
- **Emotional register**: [description]
- **Sample (most representative line)**: "[quote]" (Ch. X)
- **Sample (least representative line)**: "[quote]" (Ch. X) — potential voice break

[Repeat for each character]

## Similarity Matrix

|  | Alice | Bob | Carol | ... |
|---|-------|-----|-------|-----|
| Alice | — | 45 | 82 | ... |
| Bob | 45 | — | 38 | ... |
| Carol | 82 | 38 | — | ... |

## High-Similarity Alerts

### [Character A] vs. [Character B] — Score: [N]
- **Driving factors**: [which dimensions are too similar]
- **Evidence**: [side-by-side dialogue examples that could be swapped without noticing]
- **Recommendation**: [specific differentiation strategy]

## Voice Breaks

[Instances where a character suddenly sounds nothing like themselves — possible AI drift]

| Character | Chapter | Line | Issue |
|-----------|---------|------|-------|
| ... | ... | ... | ... |

Handling Edge Cases

Characters who should sound similar: Siblings, members of the same profession, or characters from the same background may naturally share speech patterns. Note this in the report — similarity is only a problem when characters should be distinct but aren't.

Characters with very little dialogue: Characters with fewer than 10 lines get a partial fingerprint with a confidence warning. Do not draw strong conclusions from small samples.

Dialect and accent rendering: If characters use phonetic dialect spelling, analyze the underlying patterns, not the surface spelling. "Gonna" and "going to" are the same construction at different formality levels.

Anti-Patterns

Reducing voice to gimmicks. Giving a character a catchphrase does not give them a voice. True voice differentiation comes from how characters think as expressed through how they speak — sentence structure, what they choose to say vs. leave unsaid, and their relationship to language.

Demanding perfect scores. A similarity score of 40-60 between characters in the same social group is normal. Only scores above 70 warrant concern, and only above 85 demand revision.

Ignoring narrative voice contamination. In close-third or first-person POV, the narrator's voice and the character's dialogue voice should be related. If they diverge completely, the narration may be the problem, not the dialogue.

Counting without context. Raw metrics are starting points, not verdicts. A character who uses long sentences might do so because they're evasive, or because they're precise. The interpretation matters as much as the measurement.

Prescribing voice changes that break character. Recommendations must serve the story. Do not suggest making a reserved character suddenly verbose just to differentiate them from another reserved character. Find other dimensions to adjust.