Abnormal sound detection method for compensating abnormal perception and stability by using time-frequency fusion

Abstract

The invention belongs to the technical field of abnormal sound detection methods, and particularly relates to an abnormal sound detection method for compensating abnormal perception and stability by using time-frequency fusion, which comprises the following steps of: forming perception complementation by using time domain information and frequency domain information acquired from an original audio signal x belonging to R1 * L of an acoustic target; the time-frequency domain fusion features are input into a deep neural network, state perception of a to-be-detected acoustic target is obtained through network learning, and aiming at the problem that Log-Mel spectrum features adopted by an existing abnormal sound detection method lack perception ability for a certain acoustic target, learning features constructed from a time domain angle are designed to be fused with a Log-Mel spectrum, so that the detection accuracy of the abnormal sound is improved. A win-win gain mechanism of time-frequency domain information complementation is achieved, the stability of an abnormal sound detection system can be effectively improved through the learnable features of time-frequency domain fusion designed by the method, and the problems that an existing industrial abnormal sound detection method is insufficient in stability and low in detection result credibility are solved.

Description

technical field [0001] The invention relates to the technical field of abnormal sound detection methods, in particular to an abnormal sound detection method that uses time-frequency fusion to compensate abnormal perception and stability. Background technique [0002] Anomalous Sound Detection (ASD), the purpose is to automatically identify whether the target (such as a machine or equipment) has abnormal sound, abnormal behavior or state. [0003] With the application of deep learning in the direction of audio processing, existing research provides two methods of unsupervised and self-supervised sound anomaly detection. Existing unsupervised methods learn the characteristics of normal voices by minimizing the reconstruction error and use the reconstruction error as a score to detect anomalies. Such an industrial sound anomaly detection method can provide a certain degree of abnormal sound detection performance, but its false detection rate is high and is greatly affected by ...

AI Technical Summary

Problems solved by technology

[0006] The object of the present invention is to provide a method of abnormal sound detection that uses time-frequency fusion to compensate for abnormal perception and stability, so as to solve the problem of limited performance and stability of abnormal sound detection caused by the Log-Mel spectral feature proposed in the background technology. Insufficient and existing methods deal with the problem of high multi-model complexity when dealing with different types of acoustic targets

Images