Research And Implementation Of Speech Enhancement Algorithms Based On Adaptive Filtering

Voice communication is a form of communication widely used in industrial field production scheduling,but voice signals are susceptible to interference from industrial site environmental noise,which seriously affects the clarity and intelligibility of voice signals.In order to obtain a relatively pure speech signal,a speech enhancement technique can be used to reduce the noise of the noisy speech signal.An adaptive filtering-based speech enhancement algorithm is an important algorithm in speech enhancement.This type of algorithm uses the best filtering criteria to adjust the filter weight vector in order to obtain a pure speech signal and improve speech quality.The implementation of the algorithm in this paper is based on the horizontal FIR filter structure.The related adaptive filtering algorithms are improved for two different noise types,and the improved algorithm is applied to speech enhancement.The specific research contents of this article are as follows:?1?Aiming at the Gaussian noise signal,in order to solve the contradiction between the convergence speed and steady-state error of the LMS algorithm,an improved variable step size LMS algorithm is proposed based on the existing variable step size LMS algorithm.The step size factor of the improved variable step size LMS algorithm uses a logarithmic function as the basic model.It can provide a large value at the beginning of the algorithm,which makes the algorithm converge quickly in the early stage,and a smaller value at the end of the algorithm,which makes the steady-state error of the algorithm smaller.The Matlab simulation shows that the speech signal processed by the improved algorithm is more than 10%higher than the other two similar algorithms in signal-to-noise ratio,segmented signal-to-noise ratio,speech intelligibility and PESQ,which improves the quality of speech signal.?2?For non-Gaussian noise signals with significant pulse characteristics and heavy tails,an?-stable distribution model is introduced to derive and analyze the RLLMP algorithm with better noise reduction performance.In order to further improve the performance of the RLLMP algorithm in speech enhancement,a quasi-Newton method is used to construct a positive definite matrix instead of the matrix inversion lemma to complete the algorithm iteration.Correction;the stability of the algorithm was improved,and the absolute value of the average error was reduced from 10^-1 to 10^-2.MATLAB simulation shows that the speech signal obtained by the improved algorithm has an average signal-to-noise ratio and segmental signal-to-noise ratio of more than20%compared with the speech signal obtained by the RLLMP algorithm and RLS algorithm;the average speech intelligibility and PESQ Increased by more than 5%,improving the quality of voice signals.?3?Built the hardware circuit of the dual-channel adaptive speech enhancement system with STM32F429 as the core,wrote the software,verified some functions of the two improved algorithms through the simulation function of Keil software,and completed the entire system in a realistic noise environment Test.The experimental results show that the algorithm works well in the adaptive speech enhancement system constructed,and can effectively reduce the noise of noisy speech signals.