This paper considers the scenario when a mix of signals from multiple acoustic sources is received by a microphone array. The problem is to estimate the waveform of the source of interest located in the near field of the array. The considered problem can arise in many applications such as video conferencing, acoustic room surveillance and others, when it is necessary to capture human speech against acoustic interferences. To solve this problem, an adaptive algorithm is independently applied to each narrow band of the received signal. Based on the model of interference including rank-deficient correlation matrix, a new method of robust adaptive processing is proposed. The results of numerical simulation and experiment demonstrating the robustness of the proposed method to imperfections in desired signal spatial model and to the finite sample size effect are presented.