Development Of Source Separation Algorithm In Audio Application

Nurul Amiza, Amir Hamzah (2013) Development Of Source Separation Algorithm In Audio Application. Project Report. UTeM, Melaka, Malaysia. (Submitted)

	PDF (24 Pages) Development_Of_Source_Separation_Algorithm_In_Audio_Application.pdf - Submitted Version Download (575kB)
	PDF (Full Text) Development_Of_Source_Separation_Algorithm_In_Audio_Application.pdf - Submitted Version Restricted to Registered users only Download (1MB)

Abstract

Blind source separation (BSS) is the separation signals from a set of mixed signal without the aid information about the source signals or the mixing process. Source separation problems are occurred when a number of sources is mixed and try to separate to individual source. The problem is how to construct a signal, which is able to overcome the characteristics of various sound sources. The main objective of this project is to develop and implement the algorithm of source separation using the Nonnegative Matrix Factorization 2Deconvolution (NMF2D) based on least squares function (LS). It can estimate original signals from observed signals, which consists of mixed signal and noise. By using this method, the convolutive in time and frequency is represented where it can factorize a spectrogram of music instruments corresponding to individual instruments. The music instruments can be separated using spectrogram masking based on factorization 2-de-convolution. In this project, the basic theory and literature review of this method is applied in separating audio using Matlab software. The result shows that the music instruments are piano and trumpet can be separated effectively from mixture source to original source. The proposed algorithm has applications in wide range including computational auditory scene analysis, music information recapture and automatic music transcription.

Actions (login required)

View Item