Search Effectiveness Measures for Symbolic Music Queries in Very Large Databases

In the interest of establishing robust benchmarks for search efficiency, we conducted a series of tests on symbolic databases of musical incipits and themes taken from several diverse repertories. The results we report differ from existing studies in four respects: (1) the data quantity is much larger (c. 100,000 entries); (2) the levels of melodic and rhythmic precision are more refined; (3) anchored and unanchored searches were differentiated; and (4) results from joint pitch-and-rhythmsearches were compared with those for pitch-only searches. The search results were evaluated using a theoretical approach which seeks to rank the number of symbols required to achieve “sufficient uniqueness”. How far into a melody must a search go in order to find an item which is unmatched by any other of the available items? How much does the answer depend on the specificity of the query? How much does anchoring the query matter? How much does the result depend on the nature of the repertor...

• Type: Conference
• Year: 2004
• Where: ISMIR
• Authors: Craig Stuart Sapp, Yi-Wen Liu, Eleanor Selfridge-Field
• Keywords: Diverse Repertories, ISMIR 2004, Musical Incipits, Robust Benchmarks