From October 29, 2022 to January 11, 2023
End-to-end speaker diarization models aim to solve the task of "who spoke when" without the need of separately trained modules. However, these models still struggle to deal with multiple speakers per recording. During the stay, we continued our efforts on improving their quality.
