Today, same audio file speaker diarization found 2 speakers, false

[frech87]

One week I tested Nemo speaker diarization with diar_msdd_telephonic model, and it found 4 speaker, its true. But now it found the same audio , 2 speakers. I clear cache, rebuild docker image. Not working. What happened?

[tango4j]

Hi.
I suggest to check NeMo versions, pytorch versions, and lightning versions carefully to reproduce the data.
Also checkout the latest streaming end-to-end speaker diarization model up to 4spks.
https://huggingface.co/nvidia/diar_streaming_sortformer_4spk-v2.1