Paul Primus

Paul Primus is a researcher at the Institute of Computational Perception, Johannes Kepler University Linz, specializing in audio processing and machine learning. His work focuses on sound event detection, acoustic scene classification, and language-based audio retrieval, with significant contributions to the DCASE (Detection and Classification of Acoustic Scenes and Events) challenges.

Education:

• Dr. (PhD)
• MSc
• BSc

Research Interests: Primus's research bridges audio signal processing and deep learning, addressing real-world challenges in machine listening. His work emphasizes device invariance, data efficiency, and transformer architectures for audio analysis. Key contributions include knowledge distillation for audio retrieval, multi-stage transformer training, and novel approaches to language-audio interaction. He actively explores low-complexity solutions suitable for embedded systems and edge deployment.

Publication Trends: Primus's recent work (2023-2025) shows a clear trajectory toward multimodal audio-language systems, leveraging transformers and pretraining techniques. His publications increasingly focus on data efficiency, device generalization, and practical deployment constraints, as evidenced by his DCASE challenge submissions. The integration of metadata and cross-modal alignment represents a growing research emphasis.

Activities:

• Adversarial Robustness in Data Augmentation (2020)
• Exploiting Parallel Audio Recordings to Enforce Device Invariance in CNN-based Acoustic Scene Classification (2019)

Labs and Teams: Primus is a core member of the Institute of Computational Perception at JKU, which leads research in computational audio analysis. The institute maintains strong participation in international challenges like DCASE and collaborates extensively on audio transformer development and language-audio interaction systems.