Tortoise TTS Extension for TTS-WebUI

License - the source code within this repository is licensed under the MIT license.

This extension provides a high-quality text-to-speech model with voice cloning capabilities.

Features

• High-quality speech synthesis
• Voice cloning capabilities
• Multiple quality presets (ultra_fast, fast, standard, high_quality)
• Adjustable parameters for both autoregressive and diffusion models
• Support for custom models and tokenizers
• Split prompt functionality for long text

Usage

1. Select a model (Default or custom)
2. Choose a voice from the dropdown or upload your own voice samples
3. Select a preset quality level
4. Adjust parameters as needed
5. Enter your text in the prompt field
6. Click "Generate" to create speech

Advanced Options

Model Settings

• KV Cache: Enable for faster inference at the cost of more VRAM
• DeepSpeed: Enable for optimized performance on supported hardware
• Half Precision: Enable for reduced memory usage
• Custom Tokenizer: Upload a custom tokenizer file for specialized use cases

Autoregressive Parameters

• Num Autoregressive Samples: Higher values produce better quality but slower generation
• Temperature: Controls randomness in the autoregressive model
• Length Penalty: Penalizes longer sequences
• Repetition Penalty: Discourages repetitive outputs
• Top P: Controls diversity of outputs
• Max Mel Tokens: Maximum length of generated speech

Diffusion Parameters

• Diffusion Iterations: Higher values produce better quality but slower generation
• Cond Free: Enable for better quality
• Cond Free K: Controls the strength of conditioning
• Temperature: Controls randomness in the diffusion model

This extension uses the Tortoise TTS model.