Reproduce “SynthSOD: Developing an Heterogeneous Dataset for Orchestra Music Source Separation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jaime Garcia-Martinez, David Diaz-Guerra, Archontis Politis, Tuomas Virtanen, Julio J. Carabias-Orti, Pedro Vera-Candeas
Researchers introduced SynthSOD, a large-scale, heterogeneous synthetic dataset designed to address the lack of clean multitrack data for orchestral music source separation. This dataset, generated through musically motivated annotations and high-quality synthesis, enables the training and evaluation of advanced separation systems by providing a diverse and expansive resource.
Key insights
- Musically motivated random annotations significantly enhance the diversity of synthetic data, improving model generalization across different synthetic musical expressions.
- A substantial domain gap exists between synthesized training data and real-world orchestral recordings, indicating the need for specific fine-tuning or domain-adaptation strategies for practical application.
- The complexity of orchestral music, particularly with high polyphony, presents a challenging scenario for source separation, even with large, diverse synthetic datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.