Abnormal sound detection model training method, device and computer storage medium

Abstract

The invention discloses a training method, device and computer storage medium for an abnormal sound detection model. The method includes the following steps: intercepting a sound segment with a preset duration into N sub-segments, and using H different frequency bands for each sub-segment. Sampling and filtering with a bandpass filter to obtain W sampling values, and form a three-dimensional feature tensor of N×H×W; input multiple three-dimensional feature tensors into a three-dimensional convolutional neural network for training; among them, multiple three-dimensional feature tensors Corresponding to sound clips of multiple preset durations; the sound clips of multiple preset durations include positive samples with abnormal sounds and negative samples without abnormal sounds; the loss is calculated by using a loss function that simultaneously evaluates positive samples and negative samples, and updates Parameters for the abnormal sound detection model. The invention solves the problems of inaccurate recognition and low operating efficiency in the existing abnormal sound detection.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to a training method, device and computer storage medium for an abnormal sound detection model. Background technique [0002] In the application of sound anomaly detection, the most cutting-edge method in the prior art is to train the classification network based on circular convolution through label samples, which has the following disadvantages: [0003] 1. In terms of feature extraction, existing methods mostly use hand-designed features (such as Mel cepstral coefficients (MFCC)). These methods are too dependent on the perception level of the problem and are not intelligent enough, and cannot guarantee the most suitable for the current application scenario. Excellent. [0004] 2. In terms of recognition models, existing methods mostly use circular convolutional networks and their variants as the backbone network. These models rely on the previous intermediate results in the re...

AI Technical Summary

Problems solved by technology

[0007] The main purpose of the present invention is to provide a training method, device and computer storage medium for an abnormal sound detection model, aiming to solve the problems of inaccurate recognition and low operating efficiency in the existing abnormal sound detection

Images