Research and application of environmental sound classification algorithm based on PANNs-CNN
Environmental sound classification(ESC)technology mainly involves sound feature extraction and the selection of classifier algorithms.In order to explore the best feature extraction methods and classifier combinations,this article studies and analyzes the deep learning model PANNs-CNN,and compares different feature extraction methods through experiments.The experimental results show that compared with similar models,selecting pretrained and deeper CNN models can improve the predictive performance of ESC.Log-Mel features can better preserve high-dimensional features and feature correlations of sound signals,which helps improve the accuracy of model classification.The environmental sound classification algorithm based on Log-Mel feature extraction method and PANNs-CNN14 studied in the article has the best classification accuracy on the ESC-50 dataset,and its effectiveness has been verified in practical applications.
ESCPANNsCNNLog-MelMel frequency cepstrum coefficient
NETL
NSTL
万方数据
