基于统计信号模型的卷积盲源分离.pdf

IEEE TRANSACTIONS ON AUDIO, SPEECH, AND LANGUAGE PROCESSING, VOL. 17, NO. 1, JANUARY 2009 117 An Approach for Solving the Permutation Problem of Convolutive Blind Source Separation Based on Statistical Signal Models Radoslaw Mazur and Alfred Mertins, Senior Member, IEEE Abstract— In this paper, we present a new algorithm for solving separated independently, the discrete bins usually have different the permutation ambiguity in convolutive blind source separation. scalings, and they can be arbitrarily permuted. One method to Transformed to the frequency domain, existing algorithms can ef- avoid such problems is to use a frequency-domain separation ficiently solve the reduction of the source separation problem into independent instantaneous separation in each frequency bin. How- criterion, but to restrict the time-domain impulse responses of ever, this independency leads to the problem of correctly aligning the unmixing filters to a certain maximum length [7], which these single bins. The new algorithm models the frequency-domain means that the coefficients for all frequency bins have to be separated signals by means of the generalized Gaussian distribu- modified jointly. Similar to the direct time-domain approaches, tion and employs the small deviation of the parameters between the objective function shows many local minima in which the neighboring bins for the detection of correct permutations. The performance of the algorithm will be demonstrated on synthetic algorithm can get trapped [8], and a good initialization is often and real-world data. essential to achieve good performance.