This is the code for the paper Unsupervised Topic Segmentation of Meetings with BERT Embeddings.
The code doesn't require training and uses a pretrained model from https://huggingface.co/transformers/model_doc/roberta.html See paper appendix for more information.
The entry point is eval.eval_topic_segmentation that returns the scores reported in the paper for the different methods.