Automatic Thematic Extractor

摘要

We have created a system that identifies musical “keywords” or themes. The system searches for all patterns composed of melodic (intervallic for our purposes) repetition in a piece. This process generally uncovers a large number of patterns, many of which are either uninteresting or only superficially important. Filters reduce the number or prevalence, or both, of such patterns. Patterns are then rated according to perceptually significant characteristics. The top-ranked patterns correspond to important thematic or motivic musical content, as has been verified by comparisons with published musical thematic catalogs. The system operates robustly across a broad range of styles, and relies on no meta-data on its input, allowing it to independently and efficiently catalog multimedia data.