IndexTTS is an industrial-level zero-shot text-to-speech system designed for high-quality, controllable voice synthesis, particularly excelling in Chinese language scenarios. It targets researchers and developers seeking advanced TTS capabilities, offering state-of-the-art performance and features like pronunciation correction and precise pause control.

How It Works

IndexTTS builds upon XTTS and Tortoise, integrating a conformer conditioning encoder and a BigVGAN2-based speechcode decoder. This architecture enhances training stability, speaker similarity, and audio quality. A key innovation is its character-pinyin hybrid modeling for accurate Chinese pronunciation, alongside punctuation-based pause control.

Quick Start & Requirements

• Install: pip install -r requirements.txt and pip install -e . for CLI usage.
• Prerequisites: Python 3.10, ffmpeg, PyTorch. Model weights must be downloaded to a checkpoints directory.
• Setup: Requires downloading model checkpoints (approx. 1.5GB).
• Demos: HuggingFace (link), ModelScope (link).
• Web UI: pip install -e ".[webui]" && python webui.py

Highlighted Details

• Achieves state-of-the-art performance, outperforming popular TTS systems like XTTS, CosyVoice2, and F5-TTS in benchmarks.
• Offers zero-shot voice cloning capabilities.
• Features a character-pinyin hybrid approach for improved Chinese pronunciation.
• Enables fine-grained control over pauses using punctuation.

Maintenance & Community

• Released model parameters and inference code on March 25, 2025.
• Paper submitted to arXiv on February 12, 2025.
• Community channels include QQ group (553460296) and Discord (link).

Licensing & Compatibility

• The repository does not explicitly state a license. Model weights are available via HuggingFace and ModelScope.

Limitations & Caveats

• Windows users may face issues installing pynini, requiring a conda installation.
• Contact is required for more detailed information, suggesting potential limitations in public documentation or support.