BSS Algorithm Based on Fully Connected Recurrent Neural Network and the Application in Separation of Speech Signals

Based on the traditional algorithm for blind source separation, this paper proposes a fully connected recurrent neural network algorithm for blind source separation. The self-feedback loop is increased to the algorithm. It can inhibit network into local minimum effectively; prevent the concussion; accelerate the convergence speed of weight; and applicable to nonlinear mixed situation. The simulation results show that, the algorithm has a good separation effect for multiple overlapping speech signals.