Abstract:
Music is hierarchically structured, in which the global attributes (e.g., the determined tonal structure, musical form) dominate the distribution of local elements (e.g., pitch, playing technique arrangement). Existing methods for instrumental playing technique detection mostly focus on the local features extracted from audio. However, we argue that structural information is critical for both global and local tasks, particularly considering the characteristics of Guqin music. Incorporating mode and playing technique analysis, this study demonstrates that the structural relationship between notes is crucial for detecting mode, and such information also provides extra guidance for the playing technique detection in local-level. In this study, a new dataset is compiled for Guqin performance analysis. The mode detection is achieved by pattern matching, and the predicted results are conjoined with audio features to be inputted into a neural network for playing technique detection. Advanced techniques are developed to optimize the extracted pitch contour from the audio. It is manifest in the results that the global and local features are inter-connected in Guqin music. Our analysis identifies key components affecting the recognition of mode and playing technique, and challenging cases resulting from unique properties in Guqin audio signal are discussed for further research.
