Sound scene classification method based on width and depth neural network

Abstract

The invention discloses a sound scene classification method based on a width and depth neural network, and the method comprises the following steps: firstly extracting logarithm Mel spectrum featuresfrom a sound scene audio sample, and dividing the features into a training set and a test set; designing a width neural network and a deep joint probability network; taking the logarithm Mel spectrumcharacteristics of each audio sample in the training set as input, and pre-training the two networks; constructing a joint discriminant classification tree model according to a pre-training result, and training and optimizing the joint discriminant classification tree model; and finally, inputting the logarithm Mel spectrum characteristics of each audio sample in the test set into the joint discrimination classification tree model, and identifying a sound scene corresponding to each audio sample. The constructed joint discriminant classification tree model can overcome the defects of poor generalization ability and weak stability of a single network, and the sound scene classification effect is improved by utilizing the advantage complementation characteristics of the width neural networkand the deep neural network.

Description

technical field [0001] The invention belongs to the technical field of machine hearing, relates to the width and depth learning technology, and in particular relates to a sound scene classification method based on the width and depth neural network. Background technique [0002] People's daily activities contain various sound events, and the combination of these sound events constitutes various sound scenes. Acoustic scene classification technology has a wide range of application scenarios, such as audio monitoring, multimedia retrieval, automatic assisted driving, smart home and other fields. [0003] It is difficult to improve the classification accuracy of the application of the wide neural network in the classification of acoustic scenes to a certain extent, and it is difficult to meet the practical requirements. In the past, the classification of acoustic scenes was mostly based on deep neural networks, but the long training time was the disadvantage of deep neural net...

AI Technical Summary

Problems solved by technology

[0003] It is difficult to improve the classification accuracy rate of wide neural network application in acoustic scene classification after it has been improved to a certain extent, and it is difficult to meet practical requirements.

In the past, the classification of acoustic scenes was mostly based on deep neural networks, but the long training time is the shortcoming of deep neural networks.

In fact, the wide neural network can achieve a high classification accuracy rate for certain categories of acoustic scenes, but the classification accuracy rate for other categories is low, resulting in an overall accuracy rate that cannot be increased after a certain level.

Images