Research On The Separation Algorithm Of Speech And Background Music Signals

Speech and music are two common audio signals in our lives.The mixed signal of speech and music contains a lot of useful information such as speaker identity,voice content,music melody,etc.Therefore,extracting effective information from the mixed signals of speech and music has important applications in the fields of speech recognition,audio retrieval,computer hearing,and so on.Blind source separation(BSS)technology can achieve the separation of mixed speech and musical signals.It can be understood that in the case of unknown source signal and transmission channel parameters,according to the statistical characteristics of its input source signal,only for observation The process of processing the signal to recover the source signal.In the mixed signal of speech and music,it is considered that their source signals are independent of each other.When the components of the source signal are independent,the blind source separation process is called independent component analysis(ICA).Independent component analysis is mainly composed of criterion function and optimization algorithm.The core problem when separating mixed signals is to select an optimization algorithm with superior performance to achieve the optimal result of the criterion function.Commonly used criterion functions are: mutual information minimization method,negative entropy maximization method,maximum likelihood method,fourth-order cumulant,etc.These criterion functions are used to judge the independence of each signal after separation.After the criterion function is determined,the appropriate optimization calculation is selected to perform the search to maximize its independence.Common optimization algorithms are: genetic algorithm,artificial bee colony method,particle swarm algorithm and so on.The application of intelligent optimization algorithms can overcome the optimization of the independence criterion function and enter the local optimal position.However,these traditional optimization algorithms have certain limitations.The performance of these optimization algorithms depends on the selection of control parameters.Therefore,it is proposed to use a single Pure random search of parameters for surface-simplex swarm evolution(SSSE),to overcome the impact of algorithm parameters on the performance of optimization algorithms,and to improve the effectiveness of blind separation algorithms.The idea is based on the simplex neighborhood of particles Features utilize the simplex neighborhood search mechanism and multi-role evolution search strategy.It is only necessary to set the parameter of the number of population.In the fully random two-dimensional subspace,the simplex neighborhood convex set is used to gradually approach and search for positioning to optimize,reducing the dependence on the initial value.This algorithm uses group collaborative search and competitive selection,and uses particle multi-role states in the search scheme to achieve particle diversification and improve the overall search.According to the characteristics of speech and music signals,the fourth-order cumulant is selected as the criterion function of independent component analysis.The simplex neighborhood and multi-role evolution optimization algorithm are combined with independent component analysis to blindly separate the mixed signal of speech and music.The algorithm passes The search operator's fully random searchability and multi-role state characteristics optimize the criterion function to achieve the best separation effect.Through simulation experiments,the experimental results show that the improved algorithm effectively separates speech components from background music components,and has better performance in terms of stability and separation effect.