Environmental sound recognition method based on ensemble learning and convolutional neural network

Abstract

The invention discloses an environmental sound recognition method based on ensemble learning and a convolutional neural network; and the method comprises the steps: S1, performing feature extraction:carrying out the framing and windowing of an original audio, obtaining a Mel energy spectrum of sound through a Mel filter bank, and finally obtaining a final Mel energy spectrum feature as a data set; S2, performing model training: performing model training on the data set by adopting k-fold cross validation and using a mixup data enhancement method to obtain K convolutional neural network models; and S3, performing sound testing: recognizing a sound sample to be tested through the convolutional neural network model. According to the method, k models can be trained by utilizing k-fold cross validation and k models are combined to perform voice recognition, the generalization ability of the models is greatly enhanced, and the over-fitting phenomenon is effectively relieved; in addition, for the condition of small data volume, the mixup data are used for enhancing the mixing of original samples so as to further enhance the generalization ability of the models.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, in particular to an environmental sound recognition method based on integrated learning and convolutional neural network. Background technique [0002] In the research of audio information, environmental sound recognition is an important research field, and it has great application potential in the fields of security monitoring, medical monitoring, smart home and scene analysis. Compared with speech recognition, environmental sound has characteristics such as noise-like and wide frequency spectrum, which makes the recognition of environmental sound more challenging. [0003] The existing environmental sound recognition method based on convolutional neural network usually first divides the existing data into a training set and a test set, and then uses the training set to train a model until the model converges. During the training process, the test set is used to test the model. ...

AI Technical Summary

Problems solved by technology

[0004] Existing recognition methods based on convolutional neural networks, recognition methods based on convolutional neural networks and recurrent neural networks, and recognition methods based on Gaussian mixture models all use existing environmental audio data to train a single model to identify unknown environmental audio. Identification, the models trained by this method have weak generalization ability and are prone to the disadvantage of overfitting

Images