Research On Noise Reduction And Spectrum Analysis Of Multi-channel Signal

Acoustic signal processing technology has always been one of the hot research directions in the field of signal processing.In reality,the detected target signal usually carries strong noise,which drowns the essence of the signal,which brings a lot of interference to the subsequent feature analysis.Nowadays,with the explosive growth of computer computing power and the decrease of the size and cost of sensors and detectors,multi-channel signal noise reduction has opened up a new way for noise reduction processing of acoustic signals.In this paper,the noise reduction of sound signal and the analysis of sound signal spectrum are studied.Firstly,the noise reduction of multi-channel signal is carried out,and then the characteristic analysis is carried out.Finally,the performance of the algorithm is verified by the identification experiment.(1)In this paper,an 8-channel microphone collector is designed to collect the target sound signal under the background of strong noise.In order to realize the noise reduction processing of multi-channel signals,the classic method is to use the spatial characteristics of the array signal,extract the observation data from different collectors,and then calculate the subspace matrix to realize the noise reduction.In the simulation experiment of this paper,the improved variable step size NLMS adaptive algorithm is used to reduce the noise of the clean signal with white noise.Compared with the unimproved NLMS adaptive algorithm,the signal-to-noise ratio is improved by about 10 d B on average;but the background noise is colored In the case of noise,the prior knowledge about noise is insufficient,and the noise signal correlation matrix Rn is not easy to obtain and used to filter out colored noise.Therefore,this paper combines the multi-channel noise reduction algorithm based on signal subspace with the improved VAD algorithm.Experiments In the real environment,it has achieved good results in filtering strong colored noise.Compared with the unimproved algorithm,the signal-to-noise ratio after the improved algorithm’s noise reduction processing is increased by about 10 d B on average.(2)Because the characteristic information of each target sound signal is inconsistent,this paper expounds the characteristic of each type of sound signal from the time domain waveform and spectrogram characteristics.At the same time,in order to obtain the frequency characteristics of the target acoustic signal,this paper proposes a cross-correlation peak statistical frequency feature extraction algorithm based on multi-channel signal,and the frequency feature extraction effect of the target signal is remarkable in the experiment.(3)In the last experimental chapter,design the identification experiment series signal noise reduction processing and feature analysis,in the simulation experiment and real experiment,respectively,the target signal is denoised and the frequency feature is extracted,combined with the existing data of the laboratory’s self-built database and the classifier The target acoustic signal with logarithmic mel feature is extracted for type identification experiment.Finally,the results of simulation experiments and real experiments show that,in contrast to the identification experiment of the signal group directly involved in the acoustic signal type without noise reduction processing and feature extraction signal group,the recognition rate of the signal group after noise reduction processing and frequency feature extraction is significantly improved.The recognition rate in the laboratory’s self-built database has increased by more than 30%.