livetalking

LiveTalking is an open-source real-time interactive streaming digital-human engine that drives virtual avatars with text or voice for synchronized audio-video conversations. It supports multiple digital-human models, including ernerf, musetalk, wav2lip, and Ultralight-Digital-Human, and provides features such as voice cloning, interruption while speaking, full-body video stitching, motion orchestration, and custom digital-human avatars. The system can integrate LLMs to generate responses, synthesize speech through TTS, drive real-time lip synchronization for the digital human, and output through WebRTC, RTMP, or a virtual camera. The project offers access methods such as a browser page, HTTP API, and desktop client, and includes an Avatar generation page and an admin console for uploading videos to create digital humans, monitoring session status, and adjusting global settings. Its modular architecture covers the API layer, logic layer, rendering layer, streaming layer, and plugin system. It supports TTS solutions such as EdgeTTS, GPT-SoVITS, CosyVoice, and Tencent Cloud, and can connect to large models such as Qwen or OpenAI-compatible gateways. Use cases include virtual streamers and live commerce, AI digital-human customer service, online education, intelligent voice assistants, large-screen presentations, and batch short-video production. The project provides installation instructions, Docker images, API documentation, and performance references, making it suitable for developers who need real-time digital-human conversation and live-streaming output.