Stepan Shishkin wins the Best Student Paper Award. It was also his first paper, with the topic »Active Learning for Sound Event Classification using Monte-Carlo Dropout and PANN Embeddings«, which he received at the sixth workshop on Detection and Classification of Acoustic Scenes and Events DCASE 2021.

In his paper, Stepan Shishkin presents an active learning method for sound event classification, to reduce the human workload in data annotation. Machine learning methods usually require large bodies of manually annotated data to train on. The trick is that even a few expert feedbacks significantly improve the automatic annotation of data.

The paper was written as part of our project »KI-MUSIK4.0«. The project KI-MUSIK4.0 is funded by the German Ministry of Education and Research (BMBF) (16ME0076). In the project the Fraunhofer IDMT in Oldenburg develops with renowned partners a universal microelectronic-based sensor interface for industry 4.0, which is intended to deliver added value at all levels of modern industrial system architectures. The research and development work of the Oldenburg Branch focuses on networked, acoustic sensor technology. This also includes the adaptation of artificial intelligence and machine learning methods for new AI-enabled microelectronics. This should significantly improve industrial fault and condition detection as well as automated quality control.

This paper is a great example for the excellent teamwork between the University of Oldenburg, the Cluster of Excellence Hearing4all and the Oldenburg Branch for Hearing, Speech and Audio Technology HSA. Beneath Danilo Hollosi and Prof. Dr. Simon Doclo, our former team member Dr. Stefan Goetze from the University of Sheffield was also one of the reviewers.