CS performance analysis for the musical signals reconstruction

Abstract

The Compressive Sensing (CS) method for reconstruction of musical signals is analyzed in this paper. CS is a new method for signal acquisition which has been developed in recent years. In the CS scenarios, it is possible to reconstruct the entire signal information from just a small set of randomly chosen measurements, using different minimization algorithms. Consequently, this method founds application in a large number of signal processing areas. The analyzed musical signals and the applied acquisition procedure, satisfy two important CS requirements. Namely, the observed signals have sparse representation in frequency domain, and the measurement procedure provides conservation of the main information about the signal, despite the reduction of the number of analyzed samples. Musical signals of different nature and complexity are observed in the paper. The efficiency of the CS reconstruction is analyzed for different number of available measurements. It will be shown that the minimal number of measurements required for successful signal reconstruction depends on the complexity of musical tones. Based on reconstruction error, the simple CS procedure for classification of two types of musical signals is presented. The reconstruction accuracy is measured by mean relative error between original and reconstructed signal, as well as perceptually — by listening both original and reconstructed signal.