Unser Experte Jakob Abeßer organisiert gemeinsam mit Meinhard Müller (AudioLabs) und Sebastian Stober (Otto-von-Guericke-Universität Magdeburg) die strukturierte Session »Sound Analysis für Music and Audio Signals«. Vorgestellt und diskutiert werden in dieser Session insbesondere Beiträge aus den Themengebieten des Music Information Retrieval und des Machine Listening:
The development of computational methods for analyzing, segmenting, and classifying audio signals is of fundamental importance. As in general multimedia processing, many recent advances in audio processing (including speech, music, and environmental sounds) have been driven by deep learning (DL). In contrast to traditional audio engineering approaches, which directly exploit domain knowledge in the processing pipeline, DL-based approaches attempt to learn relevant patterns from (suitably annotated and structured) training data. In this session, we want to explore the benefits and limitations of model-based and data-driven techniques for learning task-specific sound event representations. Furthermore, we want to discuss explainable DL models that can better understand sound events' structural and acoustic properties. This session invites contributions that particularly address the following topics:
- Music information retrieval (e.g., structure analysis, instrument recognition, music transcription)
- Acoustic scene classification
- Separation, detection, and localization of sound events
- Integration of domain knowledge
- Hybrid approaches for audio signal processing
- Explainable and interpretable models
- Hierarchical classification approaches
Fraunhofer-Institut für Digitale Medientechnologie IDMT