Research On Abnormal Sound Detection Algorithm Based On Psychoacoustics

The market size of headphones in China has grown rapidly,and there are defective products during the R&D and production processes of enterprises.Abnormal sound detection technology is the optimal solution for identifying faulty headphones.This thesis aims to study the implementation of abnormal sound detection algorithms for headphones and degree assessment.By selecting step frequency sweep signals as reference signals and using audio signal preprocessing to process the test signals,the reference and test signals are mapped to the perceived abnormal sound index(PASI)through the psychoacoustic model in the ITU-R BS.1387 recommendation for the perceptual evaluation of audio quality(PEAQ),which is the algorithm proposed in this thesis for assessing the degree of perceived abnormal sound(PDAS).The main research work of this thesis is as follows:1.To improve on PEAQ,which uses pure tone signals as reference signals that can only test single frequency points,step frequency sweep signals are selected to evaluate most frequency points between 20 Hz and 20 k Hz,and endpoint detection technology is used to extract effective signals,which is more accurate than PEAQ’s method of ignoring small amounts of data at the beginning and end of the test signal.2.The psychoacoustic model in PEAQ is applied to abnormal sound detection and degree assessment.Different psychoacoustic model parameters(MOV)are obtained by inputting the preprocessed test and reference signals into two perceptual models of psychoacoustic models in PEAQ,and different MOV parameters are mapped to the PASI,which corresponds to the basic and advanced versions of PDAS for abnormal sound detection and degree assessment,respectively.3.The two versions of the PDAS algorithm are practically tested and verified.The advanced version of the algorithm has higher accuracy in abnormal sound detection than the basic version,achieving 98.6% accuracy in a noise-free environment and 95.6% in a low noise environment in enterprise applications.By comparing the advanced version of the PEAQ model with the advanced version of the PDAS algorithm,it is found that the accuracy of the PEAQ model in detecting noise is very low,and it cannot meet the requirements of actual enterprise applications,proving the effectiveness of using audio signal preprocessing.The correlation between the perceived abnormal sound index(PASI)and the subjective difference grade(SDG)was found to be high through actual subjective hearing tests and advanced PDAS algorithm tests.Therefore,this thesis proposes the PDAS algorithm based on audio signal preprocessing and the PEAQ psychoacoustic model.The PASI obtained by this algorithm can accurately identify faulty headphones in noise-free and low noise environments and meet the requirements of actual enterprise applications.