Zur JKU Startseite
Institut für Computational Perception
Was ist das?

Institute, Schools und andere Einrichtungen oder Angebote haben einen Webauftritt mit eigenen Inhalten und Menüs.

Um die Navigation zu erleichtern, ist hier erkennbar, wo man sich gerade befindet.


CP researchers organize Task in the DCASE 2024 Challenge

Florian Schmid and Paul Primus host the Task "Data-Efficient Low-Complexity Acoustic Scene Classification" in the DCASE 2024 Challenge


Florian Schmid and Paul Primus, members of the CP Institute, are organizing Task 1 of the DCASE 2024 Challenge. The DCASE Challenge, which stands for "Detection and Classification of Acoustic Scenes and Events," is a highly competitive international competition that advances the field of audio analysis and understanding.

The proposed task "Data-Efficient Low-Complexity Acoustic Scene Classification" has been accepted by the DCASE community's steering committee, launched on April 1, 2024, and will close on June 15. The outcomes of the task will be presented at the DCASE Workshop 2024, öffnet eine externe URL in einem neuen Fenster, taking place from October 23 to 25 in Tokyo.

The task of Acoustic Scene Classification focuses on developing algorithms that can infer the environmental context of audio recordings from short audio excerpts, identifying scenes such as Metro station, Urban park, or Public square. The go-to solution is to learn the association between audio and acoustic scenes with supervised machine learning. While excellent classification results can be achieved with deep learning systems, they are challenging to deploy in real-world scenarios.

To enhance the real-world applicability of deep learning systems, participants are required to design systems that are (1) lightweight enough to be deployable on edge devices, (2) robust to audio captured with different mobile devices, and (3) data-efficient. The data-efficiency aspect (3) is the main focus of the proposed task and challenges participants to develop systems that learn from limited amounts of available labeled acoustic scenes while maintaining high classification performance.

The detailed task description is outlined on the official website, öffnet eine externe URL in einem neuen Fenster and the corresponding paper, öffnet eine externe URL in einem neuen Fenster.