A Sound Classification Method of Traditional Chinese Musical Instruments Based on MEL Spectrogram and Convolutional Neural Networks

This paper proposes a method that combines Mel spectrograms with Convolutional Neural Networks (CNN) for classification of sounds from traditional Chinese musical instruments. By sampling the sounds of traditional instruments and converting them into Mel spectrograms, which are then fed into the CNN for feature extraction and classification training, high recognition accuracy has been achieved. This method fully leverages the time-frequency localization characteristics of the MEL spectrogram in audio signal processing and the hierarchical feature learning capability of CNN in image recognition, significantly improving the accuracy and efficiency of instrument sound classification. Experimental results on a dataset of sounds from five representative traditional Chinese musical instruments demonstrate that the model achieved a classification accuracy of 95.04% on the validation set, while maintaining robust generalization capability on short-segment datasets.