Voice Training Script Generator

Generate reading scripts for voice cloning and TTS training data. Uses your measured words-per-minute (WPM) rate to create text that will take a specific duration to narrate.

Prerequisites

• uv - Fast Python package manager
• Gemini API key from https://aistudio.google.com/apikey

Install uv if not already installed:

curl -LsSf https://astral.sh/uv/install.sh | sh

Setup

# Set up API key
cp .env.example .env
# Edit .env and add your Gemini API key

The virtual environment and dependencies are automatically installed when you first run ./generate.sh.

Manual setup (optional):

uv venv .venv
source .venv/bin/activate
uv pip install -r requirements.txt

Get your Gemini API key at: https://aistudio.google.com/apikey

Workflow

1. Measure Your WPM (Optional)

If you want accurate timing, first measure your speaking rate:

1. Place one or more MP3 files of yourself speaking in the wpm-measure/ directory
2. Run the WPM calculator:

./calculate_wpm.py

3. Update config.json with your measured WPM

2. Generate Reading Scripts

Simply run:

./generate.sh

The script will prompt you for:

• Duration (in minutes)
• Style (select from numbered list)
• Number of chunks
• Optional topic focus

Command-line mode (for scripting):

# Basic: 30 minutes of conversational text in a single file
./generate.sh -d 30

# Specify a style
./generate.sh -d 30 -s narrative

# Split into multiple chunks (3 x 10-minute files)
./generate.sh -d 30 -c 3

# Specify chunk duration (automatically calculates number of chunks)
./generate.sh -d 30 --chunk-duration 10

# Add a topic focus
./generate.sh -d 60 -s technical -t "Python programming"

# Show help
./generate.sh --help

Available Styles

• conversational - Natural, casual tone (default)
• narrative - Audiobook-style prose
• technical - Technical explanations and tutorials
• news_anchor - Professional broadcast style
• storytelling - Immersive fiction/stories
• educational - Documentary-style informative content
• podcast - Engaging monologue format

Output

Generated scripts are saved to output/session_<timestamp>/:

output/session_20241128_150000/
├── script.txt          # Single file mode
├── chunk_01.txt        # Multi-chunk mode
├── chunk_02.txt
├── chunk_03.txt
└── metadata.json       # Generation details

Configuration

Edit config.json to customize defaults:

{
  "wpm": 198,
  "default_style": "conversational",
  "available_styles": [...],
  "default_chunk_duration_minutes": 10,
  "output_directory": "output"
}

Directory Structure

.
├── calculate_wpm.py    # WPM measurement tool
├── generate_text.py    # Text generation script
├── generate.sh         # CLI wrapper (auto-installs venv)
├── config.json         # Configuration
├── requirements.txt    # Python dependencies
├── wpm-measure/        # Place audio samples here for WPM calculation
├── user-context/       # WPM analysis results
└── output/             # Generated scripts