Time stamps in out.rttm file for speaker_diarization

[bryant0918]

I have a 5 second audio file with two different speakers one after the other. It looks like it is able to recognize the two speakers but the time stamps must be wrong because it says that speaker 2 starts at 6.40 (seconds?) and lasts for 3.88 (seconds?).

Here is the output of the actual out.rttm file:
SPEAKER Anonymous 1 0.52 5.08 <NA> <NA> Anonymous_0 <NA>
SPEAKER Anonymous 1 6.40 3.88 <NA> <NA> Anonymous_1 <NA>

And attached is the .wav file I used to evaluate.
an4_diarize_test.zip

Does anyone know why the time just seems proportionally off like it's counting in half-seconds?

[ai-nikolai]

Hi @bryant0918,

Perhaps unrelated, how is diarization done using WavLM in general? Could you point me to a script or demo?