Research On Geometric Spectrum Reduction Algorithm Based On Joint Smoothing SNR Estimation

Speech communication is the most valuable means of communication under the influence of the transmission medium.However,due to the complexity of the surrounding environment,the transmission of speech signals will inevitably be contaminated by different background noises.It is because of the existence of these non-stationary noise sources that the receiving end is difficult to perceive the transmitted contaminated speech signals.Communication processing system performance is severely damaged.The task of speech enhancement technology is to recover noise from noisy speech,optimizing the quality and intelligibility of the output speech.In recent years,speech enhancement algorithms have been widely concerned by researchers from all over the country,especially in the case of strong noise interference,which has become a research hotspot.This paper will mainly focus on the research and improvement of the estimation of core parameters.The main research contents are summarized as follows:Firstly,based on the development history and research significance of speech enhancement,according to the different salient features of speech enhancement algorithms in time-frequency domain,four widely used a priori SNR estimation algorithms are briefly enumerated,which are explained in principle and supported by simulation data.Under the condition,the key points highlight the decisive significance of the prior signal-to-noise ratio parameter to the speech processing system.Secondly,the direct judgment algorithm is analyzed,and the simulation experiment is used to prove the decisive influence of the smoothing parameter on the estimation result of the algorithm.Furthermore,the a priori SNR estimation algorithm based on the recently proposed fusion coupling factor is further analyzed and studied,and an empirical constant is introduced.The posterior signal-to-noise ratio in the likelihood estimation is recursively smoothed,which effectively reduces the systematic bias generated during the estimation process.Then,the geometric spectrum subtraction algorithm is theoretically analyzed.Because the algorithm has limited ability to track the instantaneous SNR,it is considered to improve its instantaneous the priori SNR estimation value,and introduce a coupling factor to take two different smoothing parameters.The instantaneous a priori SNR is adaptive joint smoothing,and a new geometric spectroscopy algorithm for joint smoothing SNR estimation is proposed.Finally,in order to prove the effect and quality of the improved algorithm,the program code is written by MATLAB software to simulate the experimental design.The experimental data output is valid to verify the superiority of the proposed algorithm.Finally,the basic framework and content of the full text are summarized and summarized,and the future research direction of speech enhancement is analyzed and forecasted.