On March 20, 2025, Tampere University hosted the REPERTORIUM Workshop, a focused academic event that brought together researchers in signal processing and machine learning, along with attendees from various backgrounds who share a strong interest in classical music and its technological study.The workshop was part of the Horizon Europe-funded REPERTORIUM project, which aims to develop AI-driven methodologies for the preservation and dissemination of European music heritage.
The event featured a rich program of talks and discussions, including:
- Presentation of REPERTORIUM – A comprehensive overview of the project’s aims and current developments in AI for music heritage.
- New datasets for orchestral music source separation – Introduction of SynthSOD and The Spheres datasets, designed to support supervised and unsupervised learning models in complex musical contexts.
- Score-informed music source separation – Approaches to improving generalization from synthetic training data to real-world classical recordings using symbolic alignment and score-based representations.
- Unsupervised and semi-supervised techniques – Explorations into adversarial training, domain adaptation, and domain-invariant representations for music source separation.
The workshop facilitated interdisciplinary dialogue on the challenges and innovations at the nexus of technology and cultural heritage, emphasizing reproducibility and open-access dissemination. With ongoing media coverage and public concerts, REPERTORIUM is expanding the impact of computational research in music beyond academia.
Visit www.repertorium.eu to access publications, datasets, and project updates.