BQ-Wang0511/TalkLikeYou

Official implementation of Talk Like You: Imitating How You Speak in Real-Time Talking Head Generation. Baiqin Wang, Zhixing Ding, Jijie Li, Jiankuo Zhao, Zhen Lei, Xiangyu Zhu MAIS, Institute of Automation, Chinese Academy of Sciences; University of Chinese Academy of Sciences; CAIR, HKISI, Chinese Academy of Sciences; Macau University of Science and Technology [](https://bq-wang0511.github.io/TalkLikeYou/) [](https://arxiv.org/abs/2610.06658) [](https://huggingface.co/doubi-killer/TalkLikeYou) [](https://github.com/BQ-Wang0511/TalkLikeYou-Streaming-RealTime-Demo) TalkLikeYou imitates a target person's speaking habits for real-time audio-driven talking-head generation. It uses a one-step Flow Matching Motion Generator in an 18-dimensional lip-motion space and supports either a preset habit ID or a reference video. The separate Streaming RealTime Demo provides a browser interface for live microphone animation, video dubbing, and interactive chat-driven avatars. On an NVIDIA RTX 3090, the demo achieves 30+ FPS generation throughput while using less than 4 GB of GPU memory, with high-clarity facial details. The recommended environment is Linux, Python 3.10, CUDA 12.1, cuDNN 9, and an…