Research On Abnormal Sound Recognition Algorithm And System For Family Environment

As people's safety awareness has gradually strengthened,many families with children and the elderly have installed surveillance systems.By analyzing the audio and video in the surveillance system,it can effectively help us discover events in the environment that require people's attention.At present,most home surveillance systems are based on the analysis and processing of video information,but video surveillance is easily affected by occlusion and light.The processing and analysis of audio signals can avoid these problems.At the same time,the storage space required for audio files is smaller than that of video files,which can also protect the privacy of users.Therefore,the combination of video and audio can further improve the performance of the home monitoring system.effect.There are currently two difficulties in the field of abnormal voice recognition in the home environment.One is that the abnormal voice recognition algorithm is still based on the traditional machine learning algorithm,and the traditional machine learning algorithm is capable of distinguishing unstructured sounds.Limited,the recognition accuracy rate is not high;the second is the abnormal sound recognition algorithm for the home environment.Currently,there is no system that can put it on the ground and perform real-time monitoring,nor can it verify the true validity of the algorithm.In view of the above two difficulties,this article has launched the following research:Research on the environmental sound recognition algorithm based on the neural network framework.The focus is on the environmental sound recognition algorithm based on three models of multilayer perceptron,convolutional neural network and recurrent neural network.The model structure and feature extraction method of the algorithm are introduced,and the performance of the above algorithm is tested on the Urban Sound8 K data set.Through experimental comparison and analysis,the advantages and disadvantages of the three algorithms are obtained,and the corresponding module fusion method is proposed,which provides a theoretical basis for the improved algorithm proposed later.In view of the insufficient ability of traditional machine learning algorithms to distinguish unstructured sounds,which leads to the problem of low recognition accuracy,based on the previous improvement of the neural network model,the attention mechanism and convolutional neural network are proposed.The CRNN-A model of Convolutional Recurrent Neural Network(CRNN)integration.The model architecture and key technologies of the algorithm are introduced,and the performance of the CRNN-A model is tested on the Urban Sound8 K data set.The results show that the structure of the CRNN-A model is richer,the feature utilization rate is higher,and the test accuracy rate is as high as 91.29%,which far exceeds the recognition accuracy rate of the traditional algorithm,and compared with the published LSTM model,the recognition accuracy rate is improved More than 7%.It proves the effectiveness of CRNN-A model in the field of voice recognition.This paper designs and implements an abnormal voice recognition system for the family environment,and uses the CRNN-A model as the voice recognition model in the system.Recognize the abnormal sounds in the home environment,and timely feedback the identified abnormal results to the user,remind the user to deal with the incident in time,and verify the effectiveness of the algorithm.