Spaker: Cristina Moratilla.
Abstract: Sound Event Detection is one of the most developed fields in the area of audio signal processing in the last decades. The objective of such detection is to locate the start and end instants of audio events and and classify them within a known set of classes. Throughout the last years different international evaluations have been organized, where DCASE (Detection and Classification of Acoustic Scenes and Events) stands out. The aim of this evaluation is to support research on sound event analysis methods. This Final Degree Project, carried out together with the AUDIAS research group of the Universidad Autónoma de Madrid, has been developed on one of these challenges, specifically task 4 of DCASE 2021. The baseline system provided by the DCASE organization uses a semi-supervised learning method called Mean-Teacher, based on a teacher-student scheme. Taking this system as a reference, the objective of the project is to improve performance by optimizing the model selection criteria. To this end, experiments are proposed such as evaluating the selection criterion on teacher model instead of student model, or using criteria based on PSDS (Polyphonic Sound Detection Score), a recently introduced metric for the evaluation of sound event detection systems.