1 paper
Naijun Zheng, Xucheng Wan, Kai Liu +1
In hours-long meeting scenarios, real-time speech stream often struggles with achieving accurate speaker diarization, commonly leading to speaker identification and speaker count e…