Research And Application Of Abnormal Sound Monitoring Based On Spectrum Analysis

In modern production and life,sound contains more and more information.A large amount of sound can be used to monitor abnormal events through intelligent analysis and discrimination.With the development of artificial intelligence technology,abnormal sound monitoring technology has received more and more attention and has a very wide application background.Based on the analysis and comparison of common audio retrieval methods,this paper designs and implements an abnormal sound monitoring method based on the frequency spectrum characteristics of sound.(1)A feature extraction method based on MFCC-TD fusion is proposed to describe the dynamic characteristics of the signal from the perspective of spectral characteristics;(2)In the process of data collection,abnormal sound data are often much smaller than the scene sound data,leading to the imbalance of the proportion of various samples,which requires the equalization of various samples;(3)The random forest model is used to train the audio characteristic parameters,and the final prediction category of the input audio signal is obtained to realize the monitoring of abnormal sound.In order to verify the performance and overhead of the algorithm,experiments are carried out on the collected data sets containing abnormal sounds and scene sounds.Through the comparison and analysis of the experimental results,it is proved that the method has good accuracy and real-time performance,and can still maintain good performance under the background of strong noise.Based on the above methods,this paper developed an abnormal sound monitoring system based on voice recognition technology,which mainly includes sound monitoring function,real-time query function of sensor status,data visualization function of pickup,storage and query function of historical data of pickup.The tunnel water inrush accident is taken as the application background,and the system application analysis is carried out under the condition of different signal-to-noise ratio.The results show that the system has a strong ability of water inrush sound recognition and antinoise,which can meet the requirements of timeliness,and can provide a basis for water inrush accident monitoring.