Research On Environmental Sound Classification Based On Deep Learning

The automatic classification of urban environment sound has become a key point in the process of urban informatization.It refers to the classification of different urban environment sounds,such as natural environment sound,family environment sound,road environment sound,through computerized classification,in the case of background noise,to identify and sort out the surrounding environment.Although the application of environmental sound classification in the field of great potential,but the effect is not particularly ideal,relative to voice recognition and music retrieval such hot spots,environmental sound research there are many aspects of the work can be further deepened.This paper focuses on the classification of environmental sound.The corresponding solutions are proposed,and the related methods proposed in this paper are validated by experimenting on the problem of model selection and the lack of marking data set.The main work includes the following three aspects:First,the development of environmental sound classification research and the current mainstream research methods were reviewed.Second,at present,the classification of environmental sound mainly uses the method of bottom feature extraction and unsupervised clustering.This method has the problem of limited classification accuracy.Even if some scholars use CNN for environmental sound classification with the original sound data,resulting in the calculation cost being too large.In this paper,a hybrid model of classification method based on MFCC and CNN is proposed.Firstly,the MFCC primitive feature extraction is carried out.Then,the CNN model is used to extract high-level features and then classify them.The experimental results show that the hybrid model proposed in this paper has a good effect in both computational complexity and accuracy.Thirdly,this paper proposes a four-level data expansion algorithm,which is used to expand the original data set.In this paper,we propose a four-level data expansion algorithm for the expansion of the original data set.The experimental results show that the data set expansion method proposed in this paper has obvious advantages in improving the classification accuracy of CNN network,and can get better results.