# Chatterbox Gianna premium clone voice for AutoShorts

Session learning from TrueTraceShorts / Everyday Red Flags videos.

## Approved default voice

The user accepted the female Chatterbox clone variant as the ongoing default for AutoShorts:

```json
{
  "voice_mode": "clone",
  "reference_audio": "Gianna.wav",
  "temperature": 0.80,
  "exaggeration": 0.52,
  "cfg_weight": 0.50,
  "speed_factor": 1.0,
  "pause_ms": 500,
  "style": "warm narrator, conversational, calm, clear"
}
```

## Why this changed

Taylor human pacing v2 improved earlier videos, but the user noticed some sentence-end emphasis was still unnatural. Testing audio-only variants before re-rendering was useful:

- Taylor Continuity: longer paragraph chunks, lower exaggeration.
- Taylor Warm Narrator: warmer, more human.
- Ryan/Michael sanity check: identify voice-specific cadence issues.
- Premium clone route: reference-audio clone with `Gianna.wav`.

The accepted direction was the premium female clone route, not Taylor.

## Workflow rule

For future AutoShorts voice work:

1. Use Gianna clone premium v1 by default.
2. Keep chunks conversational and avoid too many hard sentence-ending generations.
3. If cadence is questioned, render audio-only A/B samples first, then re-render video only after approval.
4. After changing voice, produce a fresh video hash and posting-package hash before upload approval.