modelscope / 3D-Speaker

A Repository for Single- and Multi-modal Speaker Verification, Speaker Recognition and Speaker Diarization

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Numbers of speakers

Tortoise17 opened this issue · comments

I have tried to test the speaker diarization, so far, for different files, it is not making detection of speakers more than 4. Is there any place where the maximum numbers of speakers are defined as threshold or this is the limitation of the engine and it is failing in this case? If you can guide me.

in config file, there is cosine threshold, and also the pval. Does it improve the results by changing the values?

In the "conf/diar.yaml", there's a parameter for setting the maximum number of speakers, "max_num_spks: 15."

You can decrease the pval to increase the number of speakers. However, the more important thing is to ensure that speech activity of each speaker in the audio lasts longer than 30s, as it can be difficult to recognize speakers with shorter durations.