EXTRACTION OF THE MELODY PITCH CONTOUR FROM POLYPHONIC AUDIO

MIREX 2005 is the second evaluation of algorithms related to music information retrieval (MIR). This document describes our submission to the MIREX audio melody extraction contest addressing the task of identifying the melody pitch contour from polyphonic musical audio. We use mainly a data-driven approach – implementing standard audio signal processing techniques like Fourier analysis, instantaneous frequency estimation and sinusoidal extraction. Nevertheless, some high level knowledge is applied. Examples of such knowledge include perceptually motivated methods (psychoacoustics, auditory stream segregation) as well as basic voice-leading principles. The result of the Mirex 2005 evaluation proves that our algorithm performs best in respect of runtime and overall accuracy, a measure which combines voicing detection (discrimination between melody and non-melody parts) and pitch estimation.