Coding of sung queries for music information retrieval

Abstract

Much research in music information retrieval has focused on query-by-humming systems, which search melodic databases using sung queries. The database retrieval aspect of such systems has received considerable attention, but the query processing and the resulting melodic representation have not been examined as carefully. Common methods for query processing are rudimentary, which suggests that more advanced methods might improve retrieval performance. Researchers have also proposed that coarsely quantized melodic representations might improve performance, but these claims have not been carefully investigated. We examine several advanced query processing methods as well as quantized melodic representations for a query-by-humming system. We compute the retrieval accuracy of a complete query-by-humming system that uses these transcription methods along with varying degrees of pitch and duration quantization. We also compare the transcription methods in isolation by computing their segmentation performance. The results show that more advanced query processing can improve both segmentation performance and retrieval performance. Further, coarsely quantizing the melodic representation generally degrades retrieval accuracy rather than improving it.