Bilibili's Index team has proudly declared the comprehensive open-sourcing of its self-developed autoregressive zero-shot text-to-speech system, IndexTTS-2.0. This system marks a significant leap forward in the practical deployment of zero-shot TTS technology. Leveraging two groundbreaking innovations—a time encoding mechanism and the decoupled modeling of vocal timbre and emotion—it effectively tackles the longstanding technical hurdles of duration management and emotional conveyance in speech synthesis.

IndexTTS-2.0 showcases remarkable versatility in generating speech, rendering it suitable for a broad array of applications, including AI voiceovers, audiobooks, animated comics, video translation, voice dialogues, and podcast creation. It offers vital technical backing for the global expansion of content, significantly reducing the obstacles for high-quality content to transcend language barriers.

At present, IndexTTS-2.0 has made available its project paper, full codebase, model weights, and an interactive online demo page. The IndexTTS team has expressed their commitment to continually refining the model's performance in the future, with plans to progressively release additional resources and tools. They aim to collaborate with the developer community to foster an open and flourishing ecosystem for speech technology.

GitHub repository: GitHub - index-tts/index-tts: An Industrial-Level Controllable and Efficient Zero-Shot Text-To-Speech System.

Paper link: [2506.21619] IndexTTS2: A Breakthrough in Emotionally Expressive and Duration-Controlled Auto-Regressive Zero-Shot Text-to-Speech.

Demo link: IndexTTS2: A Breakthrough in Emotionally Expressive and Duration-Controlled Auto-Regressive Zero-Shot Text-to-Speech.

Model download links: ModelScope/IndexTTS-2, Hugging Face/IndexTTS-2.

Online demo access: https://huggingface.co/spaces/IndexTeam/IndexTTS-2-Demo.