Separation of Each Instrumental Sound in Monaural Music Signals Applying Adaptive Filters with Statistical Least Square Criterion

Abstract

An optimal filter that separates a tone of particular pitch highness played by particular instrument applying the statistical least square criterion is discussed. The filter is based on a priori autocorrelations of and cross correlations among tones. It firstly identifies current amplitude factors of tones; then, it evaluates cross correlation between current input signal and each tone. Finally it obtains the estimate of seeking tone from the input signal which is mixture of sounds played by instruments and corrupted by white noise. Some numerical examples with violin, trumpet and flute tones are also introduced.