Reproduce “Frame-Level Multi-Label Playing Technique Detection Using Multi-Scale Network and Self-Attention Mechanism” 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
Dichucheng Li, Mingjin Che, Wenwu Meng, Yulun Wu, Yi Yu, Fan Xia, Wei Li
A research from Fudan University introduces a multi-scale network with a self-attention mechanism for frame-level multi-label instrument playing technique detection in polyphonic Guzheng music. The model achieves an 86.54% F1-score on a newly created Guzheng Tech99 dataset, outperforming baselines by over 15% in this complex multi-label context.
Key insights
- Multi-scale network architecture is crucial for effectively fusing long-range and high-resolution features, accommodating IPTs with diverse temporal durations.
- The self-attention mechanism significantly contributes to model performance by capturing extended temporal dependencies, leading to a 2.53% F1-score improvement.
- Residual connections within the network enhance learning capability and overall performance by mitigating vanishing gradients and facilitating deeper network training.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.