Coqui TTS Model Evaluation Tool

A professional tool for generating and evaluating audio samples from all English Coqui TTS models. This tool helps you identify the best text-to-speech voice for your specific use case by automatically generating samples from each available model.

🚀 Features

• Automatically fetches all available English TTS models
• Generates audio samples with customizable text
• Organizes output in a dedicated samples directory
• Real-time progress monitoring
• Detailed success/failure reporting
• Docker-based execution for consistency

📋 Requirements

• Windows with PowerShell
• Docker Desktop installed and running
• Internet connection for downloading models
• At least 2GB of free disk space for models and samples

🛠️ Setup

1. Clone or download this repository

   git clone <repository-url>
   cd mozilla-tts

2. Ensure Docker is running

   • Start Docker Desktop
   • Wait for the Docker engine to be running (green icon in taskbar)

💻 Usage

1. Open PowerShell in the project directory

2. Run the script:

   powershell -ExecutionPolicy Bypass -File ./tts_monitor.ps1

3. Monitor Progress:

   • The script will show real-time progress for each model
   • Generated audio files will be saved in the audio_samples directory
   • Each file will be named: test_tts_models_en_<model-name>.wav

⚙️ Configuration

Edit the variables at the top of tts_monitor.ps1 to customize:

$outputDir = Join-Path $localPath "audio_samples"  # Output directory
$ttsText = "This is a test of the model: {MODEL}..." # Sample text
$maxTries = 10           # Number of retries
$sleepSeconds = 2        # Wait time between retries
$audioFilePrefix = "test_" # Output file prefix

📁 Directory Structure

mozilla-tts/
├── tts_monitor.ps1      # Main script
├── readme.md           # This file
└── audio_samples/      # Generated audio files
    ├── test_tts_models_en_ljspeech_tacotron2.wav
    ├── test_tts_models_en_vctk_fast_pitch.wav
    └── ...

🔍 Evaluating Models

1. Listen to Generated Samples

   • Check the audio_samples directory
   • Each file includes the model name in its spoken text
   • Compare voice quality, clarity, and naturalness

2. Choose the Best Model

   • Consider your use case requirements
   • Note model names of preferred voices
   • Use the chosen model name in your final implementation

❗ Troubleshooting

Common Issues

1. "Docker not running" error

   • Ensure Docker Desktop is running
   • Wait a few seconds and try again

2. "No English models found"

   • Check your internet connection
   • Verify Docker can pull images

3. "Permission denied" errors

   • Run PowerShell as Administrator
   • Check folder permissions

Error Recovery

• The script tracks successful and failed generations
• Failed models are listed at the end
• You can safely re-run the script to retry failed models

🔄 Updates and Maintenance

• Check for new Coqui TTS releases periodically
• Update the Docker image tag if needed
• Keep Docker Desktop updated

📝 Notes

• Generated audio files are saved in WAV format
• Each model may take several minutes to download initially
• Models are cached in Docker for subsequent runs

📫 Support

If you encounter issues:

1. Check the troubleshooting section
2. Verify your setup matches requirements
3. Check Docker logs for detailed errors

🔒 Security Note

The script runs with execution policy bypass for convenience. Always review scripts before running them with elevated privileges.