Melody-Retrieval based on Pitch-Tracking and String-Matching Methods∗
Emanuele Pollastri · 1999
techniques, and computer tools for the preservation, the structural organization, and the intelligent query of musical audio archives stored on heterogeneous magnetic media " research, Finalized Project "Cultural Heritage " (Subproject 3, Topic 3.2, Subtopic 3.2.2, Target 3.2.1). The emergence of audio data types in databases and the rapid increase in speed and capacity of computers require new information retrieval methods dedicated to audio files. Our goal is to let users search musical pieces by melodic audio-content. We developed a pitch-tracking system based on RMS-power segmentation and harmonic gathering; this software can measure pitches from monophonic sources in the range of 50 Hz to 20 KHz, with over 90 % of successful estimates, and can set the basis for a potential polyphonic pitch recognition. Resulting pitches derived from two different audio files can be compared by means of another software tool which is able to measure their degree of melodic similarity. For this purpose, three approximate string matching algorithms and a string distance calculation algorithm were implemented; all these algorithms are based on dynamic programming. Several different representations of melodic lines are available: absolute notes, intervals, relative intervals to a given note, exact and relative rhythm. The comparison can be carried out using only notes, only rhythm or both notes and rhythm. This system returns a ranked list of matching points and their degree of similarity. Preliminary results show that it’s actually possible to achieve a concise distance-measure between melodies acquired from audio files and to realise systems for melody-matching or melody-retrieval. 1.