Efficient blind separation of convolved sound mixtures

We present an extension to recent approaches to blind source separation. Bell and Sejnowski (see Neural Computation 7, MIT Press, Cambridge, MA., 1996) proposed a robust algorithm for separating instantaneous mixtures. Extensions were proposed by Torkkola (see IEEE Workshop on Neural Networks for Signal Processing, Kyoto, Japan, 1996) and Lee et al. (See Advances in Neural Information Processing Systems 9, MIT Press, Cambridge, MA., 1997) for separating convolved mixtures but the computational overhead and the convergence behavior of these algorithms were not ideal. A frequency domain extension is presented which improves the stability and the performance of these algorithms.