Music homogeneity analysis through instantaneous frequencies

This paper describes a new method of music data retrieval. Proposed method provides a way to determine tune's internal homogeneity, which means detecting which parts of the tune sound typical to whole tune and which do not. Such information can be used for various purposes such as tune thumbnail creation or tune cleaning from untypical parts for further automatic analysis. In our method, we slice music piece into frames and compare these frames with the whole tune by means of Instantaneous Frequency Spectrum, which gives a comprehensive approximation of frames homogeneity. This approximation is then iteratively refined. The final result shows homogeneity of short frames to the whole tune. By setting different thresholds we can get desired level of accuracy. However, proposed automatic threshold shows good results throughout all experiments. Visual presentation of tune homogeneity gives a good description of a piece's internal structure.