基于自适应白化的音乐节拍实时跟踪算法*

提出一种基于自适应白化的音乐节拍实时跟踪算法。在对音乐信号进行触发点检测之前，构建音乐信号短时傅里叶变换的当前频域峰值表，对各频段的频谱幅值进行自适应加权，使各个频段保持相似的动态范围，改善音乐信号起伏变化较大时触发点检测准确性低的问题，进而提高节拍实时跟踪效果。在MIREX2006标准测试库上的节拍跟踪实验表明，自适应白化的引入可整体提高P-score，特别是对于起伏变化很大的音乐信号，节拍跟踪效果的提高非常明显。

Adaptive whitening for real-time music beat tracking

This paper proposed an adaptive-whitening-based real-time algorithm for music beat tracking, introduced adaptive whitening into onset detection phase of a music beat tracker. It preprocessed the STFT frames by adaptively normalizing the magnitude of each bin according to a present spectral peak table. This allowed each magnitude bin to achieve a similar dynamic range over time and improved onset detection performance for music with spectral roll-off effects and strongly varying dynamics. Experimental results on the MIREX 2006 dataset shows that the proposed adaptive-whitening-based beat tracker achieves superior performance and it can significantly improve the P-scores for music samples with strongly-varying dynamics.