Lightweight abnormal sound event detection method based on adaptive width self-attention mechanism

Abstract

The invention discloses a lightweight abnormal sound event detection method based on a self-adaptive width self-attention mechanism. The method comprises the following steps: firstly, performing signal processing on an audio with a label to obtain a certain time-frequency characteristic representation of the audio; secondly, the feature representation (generally a vector or a matrix) with the label is taken as input to be given to the adaptive width self-attention mechanism model, then the adaptive width self-attention mechanism model has a defined loss function and a randomly initialized attention weight, the loss value of the label is calculated according to an adaptive self-attention mechanism algorithm, and the self-adaptive width self-attention mechanism model is used as a self-adaptive width self-attention mechanism model; next, the self-adaptive attention weight is updated by using a back propagation algorithm, and updating iteration is continuously performed on three input weights of attention until a loss function reaches a minimum or ideal state; and finally, storing the weight parameter at the moment by using a lightweight method, and predicting a section of unlabeled audio by taking the weight parameter as a model, so as to quickly and accurately detect the abnormal sound event.

Description

Technical field [0001] The present invention relates to a method of achieving an abnormal overlapping event detection using a self-focus mechanism, specifically a lightweight abnormal sound event detection method based on adaptive width self-focus mechanism. Background technique [0002] Abnormal sound event detection technology is a research area identified by acoustic events, which is important in smart homes, urban road abnormal detection, fault detection, etc. There is an important application value. [0003] The sound event detection task consists primarily consisting of signal processing and machine learning models, where common signal processing modes have noise, fast Fourier Transform, FFT, Mel Frequency Cepstral Coeffic, etc. MFCC) Feature extraction, etc. [0004] Some of the methods of constructing learning models using neural networks to perform student detection methods, including models using convolutional neural network (CNN below CNN), based on cyclic neural netwo...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problems of the current Sound Event Detection (SED), the prediction model is large, the prediction speed is slow, the calculation resources are overly dependent, and it is difficult to predict in real time. Quantitative Abnormal Acoustic Event Detection Method

This method can classify and detect the abnormal sound events contained in a piece of audio. In the case of the same signal processing method, it shows a better recognition effect than that based on CRNN, and solves the problem of slow operation and parallel operation based on RNN. And using the lightweight idea, the model size is compressed with a small loss of recognition performance, so that the model can be deployed on mobile terminals or other portable devices

Images