Sound Source Separation Of Multi-voice Environment Based On Auditory Central Nervous System

In actual environment, speech signals are often disturbed by various interferencessuch as background noise and other sound source. So we can adopt speech separation toget rid of the influence of noise and other sound source. Speech separation is achallenging subject and becomes to be a popular research area in signal processing fieldin recent years. Therefore, speech separation is a very important role and significancefor the practical application of speech recognition in multi voice environment. Thispaper has studied one kind model of speech separation based on the central auditorysystem and achieved speech separation in multi voice environment.Firstly, this paper introduces current situation of speech separation research athome and abroad, and selects speech separation model which based on central auditorysystem in multiple sound source environment; then describes principle of human centralauditory system processing speech separation.Secondly, this paper describes principle of human central auditory systemprocessing speech separation; and describes dual channel speech signal simulation ofspeech separation by HRFT; and then describes feature extraction method of the twokey speech signals in speech separation, ITD and ILD, which provides theoreticalsupport and prepare the premise for the following realization of speech separation.Then, this paper combines the method of processing voice by central auditorysystem of the human brain, and establishes an integrated voice separation modelsimulating central auditory system. Voice signal is under Multi spectral analysis byperipheral auditory model, then coincidence neuron model is established to extract thefeatures of voice signal, the voice is separated in the cell model of brain inferiorcolliculus.Finally, in this paper the proposed method has been used to simulation designedand verified on the Matlab platform. The results show that in the case of two and threesound source, the similarities of the separated voice signal and the original signal are allmore than0.97. As a result, the robustness of the method for the separation of speechsignals based on central auditory system is high.