Research On Abnormal Sound Detection Algorithm In Elevator Operating Environment

As an important means of indoor transportation,elevators play an important role in people’s lives.The safe and reliable operation of elevators has become an important issue in maintaining social and public safety.With the development of voiceprint technology,abnormal sound detection technology has been widely used in the field of public safety.There may be abnormal sounds such as explosions,children crying and glass breaking during elevator operation.The sound data collected by monitoring equipment is often a mixture of abnormal sounds and elevator operating environment sounds.Compared with other public places,the elevator operating environment is more complex,and the research shows the following characteristics: the background noise of the elevator environment makes the traditional sound event endpoint detection algorithm less effective,which is characterized by the poor noise resistance of the feature data,resulting in The frame failure rate of sound events exceeds 25%;the GMM model has high requirements on the amount of training sample data.Due to the small amount of abnormal sound data that can be collected in the elevator environment,the accuracy of the trained model for abnormal sound detection is low.Therefore,this paper studies abnormal sound detection in the elevator operating environment.The main contents are as follows:First of all,in view of the lack of abnormal sound data sets in the elevator operating environment,this paper adopts the method of sound synthesis,through the elevator sound data and abnormal sound synthesis,to build a sample library of various abnormal sounds in the elevator operating environment.The test has laid the foundation;secondly,for the selection of sound features,this paper analyzes various features in the time domain,frequency domain and cepstral domain,and adopts the method of multi-feature fusion to ensure that the features can accurately represent abnormal sound signals;The problem that the sound model has a poor effect on abnormal sound detection in the elevator operating environment,this paper starts from two aspects: on the one hand,in view of the uncertainty and short-term characteristics of abnormal sound events in the elevator operating environment,in the existing endpoints On the basis of the detection algorithm,a multifeature fusion hierarchical decision algorithm is proposed to ensure that the effective features of the sound event are input into the detection model;And adjusted the structure of the GMM-UBM model,and finally realized the detection of abnormal sound in the elevator operating environment by analyzing the effect of different mixing degrees and feature dimension input of the model.By analyzing and modeling the sound data in the elevator operating environment,this paper detects and recognizes the sounds of explosions,children crying,glass breaking and abnormal metal friction sounds.By optimizing the endpoint detection algorithm and detection model,While improving the accuracy of sound data category detection,it also reduces the time complexity of the algorithm,providing a guarantee for accurate and timely triggering of elevator abnormal alarms.The abnormal sound detection algorithm studied in this paper can be applied to the complex elevator operating environment,and provides a new technical support for the elevator equipment to accurately trigger abnormal alarms.