We will be presenting two papers at WASPAA 2021!
“Crowdsourcing Strong Labels for Sound Event Detection” by Irene Martin Morato, Manu Harju and Annamaria Mesaros is a study on how to simplify the annotation process and create an objective reference annotation for an audio file using multiple annotators that perform audio tagging of short segments. The strong labels are then estimated based on the tags, given the temporal position of the annotated segments within the longer audio file.
“Joint Direction and Proximity Classification of Overlapping Sound Events from Binaural Audio” by Daniel Krause, Archontis Politis and Annamaria Mesaros is a study of source localization in binaural audio, in coarse categories, e.g. “left, far”, “front, near”, “right, near”, etc.
The datasets for both papers will be published soon. See the “Publications” and “Datasets” sections!