Analysis of musical instrument sounds by source–filter–decay model.pdf

ANALYSIS OF MUSICAL INSTRUMENT SOUNDS BY SOURCE–FILTER–DECAY MODEL Anssi Klapuri Tampere University of Technology, Institute of Signal Processing Korkeakoulukatu 1, FI-33720 Tampere, Finland anssi.klapuri@tut.fi ABSTRACT the harmonic partials which “sample” the instrument timbre at the This paper proposes a way of modelling the time-varying spectral corresponding frequencies. Joint estimation of the source signal and energy distribution of musical instrument sounds. The model con- the filter has been previously studied in speech processing, but typi- sists of an excitation signal, a body response filter, and a loss filter cally with very different assumptions [4]. which implements a frequency-dependent decay. The three parts are The proposed method is evaluated by measuring its accuracy in further represented with a linear model which allows controlling the representing 33 different musical instruments. Also, we evaluate its number of parameters involved. A method is proposed for estimating usefulness for auditory stream formation, which refers to the task of all the model parameters jointly, taking into account additive noise. separating sounds from a polyphonic signal and classifying consec- The method is evaluated by measuring its accuracy in representing utive sounds into streams associated with a certain sound source. 33 musical instruments and by testing its usefulness in extracting the melodic line of one instrument from a polyphonic audio signal. 2. SIGNAL MODEL I