Why speaker diarization DER on Ego4D dataset is so bad?

Beginners
1

According to: pyannote/speaker-diarization · Hugging Face,

the performacne of PyAnnote speaker diarization on Ego4D dataset is very bad (very high DER rate):
enter image description here

  • Is there a reason for this bad results ?
  • What is the different between Ego4D dataset and the other datasets that the performance is inferior ?