Joint Processing Algorithm For Sound Source Separation And Localization Based On Sparsity Constraint

With the rapid development of information technology,applications like teleconference,multimedia communication and virtual reality raise higher demand for audio acquisition and reproduction.Precise reproduction of sound field is required instead of the audio signal itself.As we know that the accuracy of sound field reproduction relies on whether we can accurately acquire the signal and position of each source,so it's important for us to research on better algorithms for sound source separation and localization.In the area of separation,the nonnegative matrix factorization(NMF)is the state of the art because of its simplicity and effectiveness.However,it uses few structural property of audio signal and thus has a limited performance.For localization,the beamforming and spatial sparsity model have good performance with high complexity,and these method didn't use the relativity of separation and localization.To solve the problem mentioned above,we improve the multichannel NMF separation method by employing the sparsity constraint of audio signal and propose a united algorithm for sound source separation andlocalization.Main work in this paper is as the following:1)We propose a multichannel NMF separating algorithm(SC-MNMF)based on sparse property of audiosignal by employing both the frequency sparsity constraint of NMF bases and the sparsity constraint of activation coefficients of NMF bases in factorization of spatial covariance matrix(SCM).Experimental results shows the convergency and performance improvement of our method.2)We propose an united algorithm for sound source separation and localization.We use the spatial property matrix to calculate the direction of arrival(DOA),and update separation parameters using DOAs.Experimental results shows the convergence and effectiveness of our united algorithm.Research in this paper realizes the united algorithm for sound source separation and localization based on sparsity constraint,which make it possible for precise sound field reproduction.