Speaker Idendifications not working

[cyberhirsch]

i have a 1 hour long file with 2 distinct speakers, 50% male and 50% female, yet the are not separated.

[raivisdejus]

@cyberhirsch please share some details on your system. What OS (Windows, Linux, Mac) and what GPU if any does your system have?

Does the speaker identification finish, but speakers are not identified? Or it never finishes?

Also please try some super short audio. This will let us determine if the issue is with speaker identification code as such or with how computer resources are handled.

[cyberhirsch]

yes, it was stuck for a longer time on 75%, but it fnishes. The shorter clip yields the same results.

02-26-2026_short_01.mp3

Windows 10 with a RTX3090

[raivisdejus]

@cyberhirsch Can you go to Help -> About Buzz and then select "Show Log" button. Post the log file here.

[cyberhirsch]

logs.txt

[raivisdejus]

@cyberhirsch In the log I see that the speaker identification process finished. After that you should see a "Save" button. If pushed it will create another transcription result with speaker identification markings in it

If that does not happen, please try the latest development version and if it also fails please describe what happens after you push the save button.

[cyberhirsch]

yes, but there is only speaker 0 in your screenshot and that is not the expected behavior with more than one speaker. what about step 2?

[raivisdejus]

@cyberhirsch Short answer is that current model can not distinguish among the speakers in your sample, they are too similar. I do not see any quick and easy way to fix this.

In future if we find some new diarization model this could be improved, but for now there unfortunately is no solution