Loudspeaker abnormal sound classification using auditory perception weighted by energy entropy

In order to improve the classification accuracy of loudspeaker abnormal sounds, this paper proposes a method based on time-varying specific loudness weighted by energy entropy and principal component analysis. This method simulates human auditory perception mechanism to process loudspeaker sound response signal to obtain more effective features. The human hearing system can be divided into many parallel and functionally independent conduction pathways according to frequency. The energy of the acoustic signal in each pathway can be discerned. Therefore, the time-varying specific loudness is calculated firstly to build the quantitative correlation of loudspeaker's acoustic response signals and human hearing sensations. Each sub band loudness is weighted by energy entropy to highlight the acoustic strength variation with time and frequency. Then, important features are extracted by two dimensional-principal component analysis (2D-PCA). Finally, the whale optimization algorithm-least squares support vector machine (WOA-LSSVM) is adopted for classification. Visual analysis of the extracted features shows that this method can extract loudspeaker response signal features with better discriminability. Classification experimental results show that the average accuracy of this method reached 98.4%, which is higher than the classification method based on traditional time-frequency domain statistical features. The loudspeaker abnormal sound classification method in this paper simulates human auditory perception to extract features and is able to improve classification accuracy and automation effectively.