The Research Of Speech Enhancement Algorithm Based On Improved LMS Adaptive Filtering

In the last few years,with rapid development of the speech enhancement technology,researchers have been proposed plenty of classic algorithms,such as wiener filter,spectral subtraction,kalman filter and so on.At the same time,thanks to the rapid development of digital signal processing technology,adaptive filter has attracted the attention of researchers and speech enhancement field.The least mean square(LMS)algorithm which is the most widely used algorithm in adaptive filters at present.However,due to the existence of fixed step factor in this algorithm,the convergence speed of the LMS algorithm is always in contradiction with the steady state error.Therefore,this paper conducts in-depth research on the existing problems of the LMS algorithm,improves and optimizes the LMS algorithm.In this paper,the major research contents and work are as follows:Firstly,in order to solve the problem that the convergence speed of the LMS algorithm is always in contradiction with the steady state error,this paper proposes a variable step size LMS adaptive filtering algorithm based on BA(Bat Algorithm,BA).According to the step size parameter adjustment rule of variable step size LMS algorithm,two iteration stages have been divided in this algorithm.In the initial stage,variable step size LMS algorithm is used to achieve faster convergence speed.And in the final convergence stage,BA is used to improve LMS algorithm with smaller steady-state error.The convergence performance analysis and the comparative experiment of speech enhancement is carried out in MATLAB.The results show that the convergence speed has been improved,and the steady-state error has been reduced at the time.It also has better noise reduction effect than other algorithms.But the noise reduction effect of the algorithm is not obvious in the low SNR environment,and there is still room for optimization and improvement.Secondly,in order to further reduce the noise in noisy speech signals and improve the noise reduction performance in the low SNR environment,a speech enhancement algorithm based on BA variable step size LMS adaptive filter and MMSE-LSA is proposed.Firstly,the variable step size LMS adaptive filtering algorithm based on the BA is used to preliminary reduce the noise of noisy speech signals.And then the MMSE-LSA algorithm is used to further process the noisy signals.Finally,comparative experiment and ablation experiment is carried out to verify the capability of this algorithm.The experimental results show that under the three different type of noise and compared with other algorithms,the algorithm in this paper has an average improvement of 37% in the Signal-Noise Ratio evaluation criteria,5% in the ShortTime Objective Intelligibility evaluation criteria,and 16% in the Perceptual Evaluation of Speech Quality.Finally,this paper designs a set of the audio acquisition system with STM32F205 as the microcontroller,writes the code of the upper system,the bottom system and debugs the system.The system is used to collect human voice signal and noise signal in the real environment,and the collected voice signals are used as test data sets for comparative experiments to further testify the capability of the algorithm.