ABSTRACT
This chapter examines a package of new tools for retrieving similar polyphonic music sequences from audio collections. It also examines high performance methods for feature extraction and nearest-neighbour retrieval for polyphonic audio recordings. In the spectrum of applications of machine listening there is a wide range of audio matching problems. Based on LSH, the tool statistically divides a feature space into regions of similarity, and exhaustive searching occurs only within those narrow regions. Most of the literature on audio similarity adopts a 'bag-of-features' approach, such as that used successfully by genre-recognition systems. Apocrypha in audio recordings are those recorded performances that are falsely attributed to an artist which are, in fact, performed or composed by an artist other than the person named in the documentation. The chapter explores a low-level audio tools conforming to the MPEG-7 International Standard for extracting musically salient information from audio signals.
