Research On Speech Noise Reduction Processing Technology

With the development of science and technology,speech recognition system has been widely applied.Generally speaking,all the speech recognition systems currently need speech de-noising,and there is a speech de-noising system.Therefore,the speech de-noising algorithm plays an important role in speech recognition.In order to enhance the effect of the speech de-noising algorithm,the present speech de-noising algorithm is improved.First,the related principles of speech signal generation,the characteristics of speech signal,the algorithm of noise related knowledge and the characteristics of human ear are introduced.At the same time,the objective and subjective evaluation methods of speech de-noising are introduced respectively.For the evaluation of speech de-noising algorithm,we need not only consider the operability and accuracy of the algorithm evaluation,but also avoid the one-sided evaluation of the algorithm.After comparison,in view of the subjective and objective evaluation methods,we choose the two methods of segmental signal to noise ratio and waveform diagram by analysis and comparison as the objective and subjective methods for the evaluation of the algorithm.On the other hand,the running time of speech de-noising algorithm is also an important factor,so we also consider the time factor.Then,it is mainly aimed at improving the traditional spectrum subtraction and implementing the improved algorithm.At the beginning,the noise estimation of spectrum subtraction algorithm is partly improved.The traditional noise estimation method is based on the minimum statistics and the optimal smoothing.In the phase of the noise estimation,the noise estimation algorithm is used to estimate the noise effectively.Secondly,aiming at the characteristics of spectrum subtraction,we have implemented the method of speech noise processing.Finally,the effectiveness of the algorithm is illustrated by simulation experiments.Using the adaptive weighted average filter,adaptive weighted average filter is widely used in many fields,which is easy to produce noise in the process of noise processing.Similarly,it can be applied to noise cancellation of speech signals.There are many algorithms for adaptive weighted average filtering.In this paper,we use the LMS algorithm.On the basis of understanding the adaptive filtering algorithm and its application,the LMS algorithm is analyzed deeply,and then the basic algorithm of adaptive weighted average filtering is realized,which effectively reduces the noise of speech processing.Finally,we propose the estimation of parameters based on the AR model,and analyze and study the speech enhancement algorithm combined with the theory of Kalman filtering.The research content of this paper is the research of Kalman speech enhancement algorithm based on AR model.In view of the shortcomings of the traditional speech enhancement algorithm,this paper improves the adaptive estimation of spectral gain,and obtains the Kalman speech enhancement algorithm based on the AR model.The estimation of the AR spectrum envelope of the speech signal is modified,and the effect of speech enhancement is obviously improved.Finally,the performance index of the algorithm is evaluated by subjective evaluation and objective evaluation,and the evaluation results and the analysis of the results are given.The experimental results show the effectiveness of the algorithm.