Reproduce “Playing Technique Detection by Fusing Note Onset Information in Guzheng Performance” 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, Yulun Wu, Qinyu Li, Jiahao Zhao, Yi Yu, Fan Xia, Wei Li
Researchers developed an end-to-end system for detecting playing techniques in Guzheng performances, which integrates note onset information to refine technique localization. The system achieved an 80.76% note-level F1-score on a diverse real-world recording dataset, substantially surpassing prior approaches and demonstrating robust generalization.
Key insights
- Fusing note onset information with frame-wise IPT probabilities is critical for accurately delineating note boundaries and refining note-level technique classifications, significantly improving detection performance.
- Fully Convolutional Networks (FCNs) are well-suited for frame-level IPT prediction in variable-length audio, outperforming standard CNNs for this task.
- Training models on diverse sound sources and evaluating them on entirely separate real-world recordings is essential for developing IPT detection systems with strong generalization capabilities to unseen performance conditions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.