Phase Information Combined Decision-directed Estimator For A Priori SNR

In nowadays society,speech communication is not only limited from people to people,but also reflected in the communication between people and machines.Especially in the field of artificial intelligence and digital communications which have higher social attention,they all need to analyze,recognize and identify the speech signal through the machine.However,in the process of speech signal processing,it is often affected by environmental noise,which makes the quality and intelligibility of speech become worse and seriously.And it affects the processing quality of speech signal.In order to improve the quality and intelligibility of damaged speech and eliminate the influence of background noise as much as possible,it is necessary to pre-process the impaired speech signal before the analysis,recognition and identification.In recent years,with the continuous development of human-computer interaction technology,speech enhancement problem has got the focus attention of scholars at home and abroad.Based on the estimation of core parameters in the field of speech enhancement,the research work of this paper mainly includes the follows:First,this paper raises and expounds the basic principles of several common speech enhancement algorithms and its performance evaluation measure,especially on the prior SNR parameter system's important position in the speech enhancement algorithm.At the same time,this paper gives several common estimation algorithms of the prior SNR,and analyzes its design principle,advantages and disadvantages.Secondly,this paper analyzes the over-dependence of gain factor selection in the estimation algorithm of two-step noise elimination.By using the Gaussian distribution model of speech and noise signal,a new two-step noise reduction algorithm in this paper based on magnitude-squared spectrum estimation is proposed.The algorithm uses the estimation results of decision-directed method to obtain the prior SNR estimation of the current frame directly,thus effectively avoiding the dependence of the algorithm on the gain factor.Simulation experiments in various noise environments show that the proposed algorithm based on the magnitude-squared spectrum of a prior SNR has a better estimation effect.Then,the paper detailed analysis the decision-directed algorithm for the current frame prior SNR estimation,reveals the implication for the speech and noise phase information of unreasonable assumptions,and analyzes the assumptions for the algorithm to estimate the effect of impact.By analyzing the geometric relationship between the amplitude spectrum and phase of noise,clean speech and noisy speech signal,we integrate phase information into the estimation process of the current frame priori SNR,and put forward a new decision-directed prior SNR estimation algorithm.,The estimation step and block diagram of the proposed algorithm are presented in this paper,and verified the excellent performance of the algorithm under various noise background and input SNR.Finally,the paper summarizes the whole thesis and gives a prospect about future development direction of the estimation of the prior SNR.