Untersuchung von Melodiesuchsystemen sowie von Verfahren zu ihrer Funktionsprüfung (English shop)

Preview

Table of Contents, Datei (48 KB)
Extract, Datei (100 KB)

Owing to the flood of digital pieces of music on the Internet and on data storage media, the search for individual pieces of music is currently gaining greatly in importance. A mobile playback device such as the Apple iPod, for instance, readily allows the storage of 5,000 tracks. Until now, individual pieces of music can be selected from such a volume of data only by entering the title or the performer. A typical problem, however, is that the person searching for music does not have this information to hand, but can only remember the melody.

In technical terms, locating individual pieces of music within a large quantity of tracks is a search in a music database. If the searcher knows only the melody, it must consequently be possible to enter it directly into the search system. This requirement is met by a query-by-humming system (QBH system): the melody can be hummed to the system, and a number of similar melodies from the music database are presented as the search result.

This dissertation explains the operating principles of QBH systems and investigates how the individual system components can be tested for functionality. In order to provide a well-founded assessment of these functions, the most important fundamentals are set out. A chapter on music and melody explains the concept of melody and the possibilities for describing melodies. An overview of existing melody and music search engines is then given. A dedicated chapter considers relevant multimedia standards. The tasks of signal processing are presented in detail in the chapters on monophonic and polyphonic transcription. Methods of melody comparison are dealt with in the following chapter. A particular feature of the present work is the detailed investigation of the statistical properties of the melody databases used in QBH systems.