livetalking

LiveTalking is an open-source real-time interactive streaming digital human engine that enables text or voice-driven synchronous audio-video conversations with virtual avatars. The project supports multiple digital human models, including ernerf, musetalk, wav2lip, and Ultralight-Digital-Human, and offers features like voice cloning, interrupted speech, full-body video stitching, action choreography, and custom avatar creation. The system combines LLM-generated responses with TTS synthesis to drive real-time lip-sync for digital humans, with outputs via WebRTC, RTMP, or virtual cameras. LiveTalking provides browser pages, HTTP API, desktop clients, avatar generation pages, and management backends for video uploads, session monitoring, and global configuration. Its modular architecture covers API, logic, rendering, streaming, and plugin systems, supporting EdgeTTS, GPT-SoVITS, CosyVoice, Tencent Cloud, and other TTS solutions, as well as integration with Qwen, OpenAI, and other large models. Applications include virtual hosting, AI customer service, online education, smart voice assistants, digital signage, and bulk video production. The project offers installation guides, Docker images, API documentation, and performance references, catering to developers needing real-time digital human conversations and live streaming.