In this paper, we propose a new method of real-world blind separation for non-stationary signals (e.g.,speech signal, music). The proposed method is implemented in the time-frequency domain and achieves real-world blind separation by minimizing a cost function with the algorithm derived using the natural gradient under a given condition. We carried out computer simulations to demonstrate the validity of our proposed method. Moreover, we apply our method to an experiment which extracts two source signals from their mixtures observed in a normal room with computer and air conditioning noises. The results show that our method can achieve real-world blind separation.