In this paper, the problem of estimating the directions of arrival (DOAs) for an unknown number of sound sources is addressed using a microphone array. In outdoor environments, the reverberation is not considered, so the pure delay mixture model is typically adopted here. Since the sound signal is usually wide-band and non-stationary, the Short Time Fourier Transform (STFT) is employed. However in practical applications, the missing or appearing of sources frequently occurs, resulting in a time-varying number of targets. To deal with this problem, a new preprocessing method is presented to partition the whole time domain into some small time windows. In each time window, the sample covariance is computed and used to produce the DOA estimate. Moreover, two methods (Eigenvalue-based and Information Theory-based) are used for estimating the number of independent sources. Some selected simulation results are given to demonstrate the estimation performance for the DOAs, and the time accuracy for detecting the change of the number of sources.