Abstract: In this paper we present CompariSong? - a resolver of the query by humming task for a real audio information system. It compares a database of popular songs with a hummed-by-user melody and returns the closest matches. The proposed system performs audio indexing based on perceptual characteristics of the songs rather than the hitherto applied approach that uses manual metadata like text tags. We extract low-level audio features and transform them into a tangible indexing format. Then searching mechanism is applied that is based on comparison between the shape of the time series of extracted features. CompariSong? is a step towards intelligent audio analysis as it uses a comparison criteria derived from natural human hearing.

Key words: audio indexing, Query by Humming (QBH), audio features extraction, feature transformation, data mining

ACM Classification Keywords: H.5.5 Information Systems - Information Interfaces and Presentation - Sound and Music Computing

Link:

COMPARISONG: AUDIO COMPARISON ENGINE

Georgi Dzhambazov

http://foibg.com/ibs_isc/ibs-11/ibs-11-p04.pdf