"CGBT: Chinese Guzheng Beat Tracking Dataset"

"The Chinese Guzheng Beat Tracking (CGBT) dataset is a novel audio corpus designed to evaluate beat tracking models on complex, non-Western rhythms. It comprises 276 solo tracks of the Guzheng, a traditional Chinese plucked zither, totaling 9.1 hours of high-fidelity audio. Unlike standard Western-centric benchmarks, CGBT features highly expressive timing, extreme tempo fluctuations (rubato), and unique non-percussive acoustic properties. Specifically, distinctive performance techniques such as tremolo and pitch gliding introduce prolonged harmonic decays and drastic spectral modulations, which severely disrupt conventional onset detection. To ensure reliability, expert musicians have manually annotated precise timestamps for both beats and downbeats through a rigorous cross-verification protocol. This dataset provides a challenging new benchmark for testing algorithmic robustness and cross-cultural generalization in hierarchical metrical analysis"