PAREDA: A Multi-Accent Speech Dataset of Natural Language Processing Research Discussions