Harmonic Harmony: Advancing Musical Instrument Classification through CNN-SVM

This research intends to improve the field of Music Information Retrieval (MIR) by creating and comparing a Convolutional Neural Network (CNN) and Support Vector Machine (SVM) approaches for the precise identification of five distinct musical instruments. The process of categorizing is quite intricate because of the subtle timbre characteristics coming from the musical sounds. We have done a detailed study evaluation of the model. To evaluate each class of instrument, we have employed precision, recall, F1-score and accuracy metrics. The precision for the model in Class 1 was 89.5, the recall was 93.76, the F1 score was 91.58, and the accuracy was 97. Class 2 precision was 91.12%, recall was 94.2%, F1 score was 92.64%, and accuracy was 97%. The third class performed outstandingly with a precision of 93.28%, recall of 97.71%, Fl-score of 95.44%, and accuracy of 98%. The precision rate of class 4 was 94.53%, the recall rate was 86.18%, the F1-score was 90.16%, and the accuracy rate was 96%. Class 5 got 95.38% precision, 92.94% recall, 94.14% F1 score, and an accuracy of 98%. In order to evaluate the overall effectiveness of the model, we calculated the macro average (precision: The results displayed precision (91.63%), recall: 92.96%, F1-score: 92.79%), weighted average (precision: 92.85%, recall: 92.77%, F1-score: 92.74%), and micro average (92.77%). These measures give an overall picture of the model’s reliability. The confusion matrix gives another proof, and the fact is that the model accuracy is 92.77033344%. This demonstrates the potential of the CNN-SVM technology for a highly accurate instrument classification across various musical styles.