This paper presents a novel technique for blind source separation (BSS) of anechoic speech mixtures in the underdetermined case. A demising algorithm that exploits the sparsity of the short time Fourier transform (STFT) of speech signals is proposed. The algorithm merges constrained optimization with ideas based on the degenerate unmixing estimation technique (DUET) (O. Yilmaz and S. Rickard, 2004). Thus, the novelty in the proposed approach is twofold. First, the algorithm utilizes all available mixtures in the anechoic scenario, where both attenuations and arrival delays between sensors are considered. Second, it is demonstrated that lq minimization with q < 1 outperforms the standard choice of q = 1. Experimental results on both synthetic and real mixtures indicate significant performance gains over other BSS algorithms reported in the literature