Right ASR Wrong Speaker diarization

[liu6381810]

Hello, I used the script to perform speaker diarization on a segment of audio, and the resulting content has relatively accurate timings and text. However, there were originally five people speaking in the audio, but this method only detected one speaker, with all content labeled as Speaker 0. I would like to ask if there are any parameters that could be adjusted to optimize this result.
python3 diarize.py -a input.mp4 --whisper-model faster-whisper-large-v3

[MahmoudAshraf97]

There are known problems with long audios in diarization, the current solution is to split it up

[liu6381810]

There are known problems with long audios in diarization, the current solution is to split it up

How long is it generally appropriate to segment?
If in an audio segment, the first half is a conversation between person 1 and person 2, and the second half is a conversation between person 2 and person 3, how can I determine the time when the conversation between person 2 and person 3 starts using segmentation?

[MahmoudAshraf97]

1 hour is tested to work well, just split it every 1 hour

[liu6381810]

1 hour is tested to work well, just split it every 1 hour

As I mentioned above, for a test video that is only 10 minutes long, the automatic speech recognition (ASR) content is relatively accurate, but the speaker identification (speaker ID) is not very accurate.