Eardraft

EarDraft: Text-to-Listening Transformation

Transform reading-oriented prose into listening-optimized text that works perfectly when read aloud with flat, neutral vocal delivery (TTS, podcasts, audiobooks).

Core Principle

When text will be read aloud without pacing, intonation, or gesture, the only instruments available are syntax, segmentation, and auditory clarity. Transform text to preserve intellectual rigor while ensuring comprehension through sound alone.

Listening-optimized text should sound like a knowledgeable person talking to intelligent peers about something important. Not like a motivational speaker, not like a sales pitch, not like someone talking to children.

Quick Start

When a user asks to transform text for listening:

1. Check whether the source is legal material: a brief, opinion, statute, regulation, contract, memo, client letter, CLE script, or student outline. If so, read resources/legal-audio.md before transforming — legal prose carries meaning in citations, italics, footnotes and subsection lettering that has no audio equivalent, and the general rules below do not address any of it.
2. Determine language and format: Ask if not clear (default: English, plain text)
3. Apply core transformations: Use syntax as rhythm, control density, maintain clarity
4. Follow anti-patterns: Never fragment into staccato, never adopt motivational cadence
5. For SSML/tagged output: Use tags sparingly (2-5 per paragraph max)

Legal material

Legal sources get their own pass. Load resources/legal-audio.md and apply it alongside the language reference.

One rule from that file overrides everything else in this skill: quoted authority is never rewritten. The transformations below recast sentences for the ear. Applied to a block quote from an opinion, a statutory provision, or an operative contract term, recasting produces a misquotation — which in legal work is a professional problem, not a stylistic one. Cut by ellipsis, frame verbally, or paraphrase openly and label it as paraphrase. Never smooth a quotation silently.

Language-Specific Guidelines

Each language has culturally and linguistically optimized transformation patterns. Load the appropriate reference file:

• English: resources/english.md (default)
• French: resources/french.md - Discourse particles, melodic phrasing
• Spanish: resources/spanish.md - Regional variants, conversational markers
• Italian: resources/italian.md - Melodic rhythm, emotive connectors
• German: resources/german.md - Formal/conversational registers, compound rhythms
• Portuguese: resources/portuguese.md - Brazilian focus, discourse particles

And, orthogonal to language, load whenever the source is legal:

• Legal material: resources/legal-audio.md - citations, symbols, quotation boundaries, footnotes, register by use case

Always load the language-specific reference before transforming unless using English with familiar patterns.

Output Format Options

1. Plain Text (default)

Clean prose without markup. Use syntax and structure for pacing.

2. ElevenLabs with [AudioTags]

Inline tags for precise TTS control:

• [pause:short] (0.5s), [pause:medium] (1s), [pause:long] (2s)
• [emphasis] for critical words
• [speed:slow/fast/normal] for pacing
• CRITICAL: Use 2-5 tags max per paragraph. Over-tagging creates robotic speech.

3. ElevenLabs Implicit

No visible tags. Use punctuation for pacing:

• Periods: 0.5-1s pauses
• Em dashes: 1-2s dramatic pauses
• Paragraph breaks: 2-3s section pauses

4. SSML - Amazon Polly

<speak>
  <break time="1s"/>
  <emphasis level="moderate">critical term</emphasis>
  <prosody rate="slow">complex section</prosody>
</speak>

5. SSML - Google Cloud TTS

<speak>
  <break strength="medium"/>
  <emphasis level="strong">key point</emphasis>
  <say-as interpret-as="date">2025-11-03</say-as>
</speak>

6. SSML - Microsoft Azure TTS

<speak>
  <mstts:express-as style="newscast">
    <break time="500ms"/>
    <emphasis>important</emphasis>
  </mstts:express-as>
</speak>

For all SSML/tagged formats: Wrap only strategic moments. Most text should be clean prose.

Core Transformation Techniques

Syntax as Rhythm

Mix sentence lengths naturally. Short sentences for emphasis, longer ones for reasoning. Include dependent clauses and natural transitions.

Auditory Cohesion

Repeat key terms for reinforcement. Varying synonyms weakens recall in listening contexts.

Transparent Logic

Use clear connectives: "Next," "But that view fails," "Now consider," "The result is this."

Self-Refreshing Arguments

Reiterate thesis every few paragraphs. Listeners cannot glance backward.

Short Referential Distance

Avoid pronouns requiring recall of earlier paragraphs. Use micro-redundancy strategically.

Imageable Language

Prefer concrete nouns and active verbs. Listeners must visualize without tonal help.

Controlled Density

Clarify complex reasoning through explicit enumeration or sequencing.

Monotone Robustness

Eliminate clauses whose meaning depends on inflection. Rewrite into explicit declaratives.

Critical Anti-Patterns (NEVER Do These)

Anti-pattern: Staccato Fragmentation

DON'T: "The court considered factors. Three factors. In reaching its decision." DO: "The court considered three factors in reaching its decision."

Anti-pattern: Motivational Speaker Cadence

DON'T: "You can do this. You will succeed. The future is yours. Act now." DO: Natural conversational variety with substance over hype.

Anti-pattern: Over-Prescriptive Commands

DON'T: "Network actively. Build your portfolio. Apply now." DO: "You might consider networking, building a portfolio, or applying to positions."

Anti-pattern: Emotion Projection

DON'T: "You are excited" or "You are inspired" DO: Offer information; let listeners decide how they feel.

Anti-pattern: One-Word Emphasis Patterns

DON'T: "Think contracts. Think torts. Think procedure." DO: Natural phrasing that respects intelligent audiences.

Anti-pattern: False Intimacy

DON'T: Presumptuous "you" statements claiming to know listener's motivations DO: Respectful communication with intelligent peers.

Anti-pattern: Transforming Information into Sales Copy

DON'T: "At the forefront of a transformative shift," "your secret weapon" DO: Maintain objectivity appropriate to source material.

Anti-pattern: Over-Tagging Audio

DON'T: Tags after every sentence, emphasis on every keyword DO: 2-5 strategic tags per paragraph maximum.

TTS Markup Philosophy (CRITICAL for Tagged/SSML Output)

Natural speech rhythm comes from good writing, not excessive markup.

Pause Tags - Use ONLY For:

• Major topic shifts between distinct sections (2-4 max per paragraph)
• Dramatic effect at strategic moments
• Breathing room after dense technical content
• NOT after every sentence or clause

Emphasis Tags - Use ONLY For:

• Truly critical words carrying the entire argument (1-2 max per paragraph)
• Technical terms being defined for first time
• Crucial distinctions or corrections
• NOT every important point or keyword

Good vs Bad Tagged Example

Wrong (over-tagged, mechanical, exhausting):

As students [pause:short] in the Fall 2025 course, [pause:short] you are at the forefront [pause:short] of a transformative shift. [pause:medium]

Right (natural flow with strategic tags):

As students in the Fall 2025 Large Language Models for Lawyers course, you're entering a profession in the middle of transformation. [pause:medium] Dean Tucker's email highlights something concrete—a surge in AI-focused roles at major firms. These aren't theoretical positions. They're [emphasis] hiring now.