kaldi sre (speaker recognition) with pretrained model

[Furcifer]

I have tried few x-vector-based approaches in PyTorch/TensorFlow but unfortunately, they are not doing very well in the NIST SRE challenges.

I want to test few pre-trained models from kaldi to decide if I should use kaldi to train my models, besides the support for PLDA and another adaptation techniques is not good in any python library.

I have downloaded SRE16 x-vector pretrained model from http://kaldi-asr.org/models/m3

I have also the SRE16 evaluation set. My question is how can I evaluate the pre-trained model without training it or the training data. I didn't find any script or any detailed description on how to use the pre-trained model for inference only.