A multi-source time-delay estimation algorithm based on iterative cross-spectrum weighted histogram is studied for real-time passive direction finding of multiple sound sources with a widely spaced microphone array. By using the short-time spectral sparseness and orthogonality assumption of audio signals as well as the frequency-varying characteristic of delay ambiguity periods, instantaneous maximum signal-to-noise ratio peaks on the true delays of concurrent sounds can be obtained simultaneously without obvious sidelobes caused by phase-difference wraparound ambiguity. As a result, the common limitation in most existing sparseness based methods on the microphone spacing that must be no greater than half the minimum wavelength of signals is removed, leading to array systems being able to have both large aperture and low complexity. Simulation results verify the effectiveness of the studied technique.