Throughout the pandemic, numerous professionals who previously worked in offices have turned to virtual conferencing solutions. As such, several automatic meeting transcript solutions have emerged. While meeting transcripts can be useful, they may be difficult to navigate (especially for longer meetings). Participants often want terser overviews. Although some methods for abstractive meeting summarization exist, they usually require multi-stage ML pipelines (e.g., with template generation, sentence clustering, etc.), which cannot be optimized end-to-end. To address this limitation, Zhu et al. propose a Hierarchical Meeting summarization Network (HMNet), which applies an encoder-decoder transformer architecture to summarize meeting transcripts. HMNet, which is pre-trained on news article data, also uses a hierarchical structure for computational efficiency and a role vector to represent each speaker.