Exponentially decaying time-recursive blind deconvolution algorithm for speech dereverberation

We present a time-recursive fast algorithms for adaptive blind deconvolution as applied to nonstationary speech dereverberation. The algorithm has a computational complexity of 30L+3 multiplications and divisions per recursion (MADPR) and uses an exponentially decaying data window, where L is the length of an acoustic impulse response (AIR). Besides the adaptive property, the time-recursive algorithm can also handle the so-called "tail" problem.