ElevenLabs Voice Performance: cheerful Indian-English female, 20s in whispered conspiratorial Tone

You are writing a text-to-speech script optimized for ElevenLabs (v2 / v2.5 / Multilingual v2 models). The output will be run through ElevenLabs to produce a voice performance.

## Brief
- **Voice profile:** cheerful Indian-English female, 20s
- **Emotion:** whispered conspiratorial
- **Duration target:** 3 seconds
- **Use case:** YouTube background bed

## ElevenLabs Prompt Engineering Rules

### 1. Use punctuation as a conductor
ElevenLabs reads punctuation almost literally:
- **Period** = long beat
- **Comma** = short beat
- **Ellipsis (...)** = trailing, thoughtful pause (use sparingly — overuse causes drift)
- **Em-dash (—)** = sharp mid-sentence pivot
- **Question mark** = rising intonation (works on v2 and later)
- **Exclamation** = raised volume, but v2 is subtle — pair with word choice

### 2. Emotion tags (v2.5+ "emotion control")
ElevenLabs does NOT honor inline stage directions like `[laughing]` in the text itself — it will read them aloud. Instead:
- Adjust the **Stability slider** (lower = more emotional variance).
- Adjust the **Style slider** (higher = more expressive, more drift risk).
- Write the *emotion into the words themselves*. A whispered conspiratorial delivery comes from word choice and sentence rhythm, not bracket tags.

### 3. Phonetic hints for hard words
If the script contains a proper noun or technical term the model gets wrong:
- Spell it phonetically in parentheses the first time: "SQLite (sequel-lite)".
- Or break it with hyphens: "Anthropic" → "An-thropic".

### 4. Pacing markers
Use line breaks to insert real pauses. Two blank lines between sentences creates a dramatic beat. This is the cleanest control.

### 5. SSML (only on API)
If the user is using the ElevenLabs API (not the web app):
- `<break time="1s"/>` inserts a precise pause
- `<emphasis level="strong">word</emphasis>` can work but is inconsistent
- SSML is stripped in the web UI — do not include unless API-only

## Voice Matching
For a **cheerful Indian-English female, 20s**, recommend:
- **Model:** [Eleven v2.5 for English, Multilingual v2 for non-English accents]
- **Stability:** [suggest a value 0-100 — lower = more emotional]
- **Similarity boost:** [suggest 70-90 for cloned voices]
- **Style exaggeration:** [0-30 for natural; 40-70 for dramatic whispered conspiratorial]
- **Speaker boost:** on

## Output Format

**A. Production Script**
A 3 seconds script in plain text. No bracket tags. Use punctuation and line breaks as pacing.

**B. Voice Settings Panel**
```
Voice: cheerful Indian-English female, 20s match (suggest 2 ElevenLabs library voices that fit, OR describe for cloning)
Model: [v2.5 / Multilingual v2]
Stability: [0-100]
Similarity boost: [0-100]
Style: [0-100]
Speaker boost: ON
```

**C. Delivery Notes**
3-5 bullet points the user should keep in mind when reviewing the output (common artifacts: mispronounced proper nouns, over-emphasis on long sentences, uncanny breaths on whispered whispered conspiratorial).

## Strict Rules
- Do not put stage directions inside the script body.
- Do not use all caps for emphasis — ElevenLabs will interpret as shouting randomly.
- Do not use more than two ellipses per paragraph.
- For a YouTube background bed, keep sentences under 25 words for clean delivery.
- Write the script at reading-level appropriate to the cheerful Indian-English female, 20s (a gritty 50s British male ≠ cheerful 20s Indian-English female in vocabulary).

Now produce the full output for a whispered conspiratorial cheerful Indian-English female, 20s delivering a YouTube background bed of 3 seconds.