Publication

Abstract : Musical instrument recognition is a challenging task with applications in music information retrieval, audio processing, and automated transcription. This study presents a Convolutional Neural Network (CNN) model leveraging Mel spectrograms from the IRMAS dataset to classify 11 instrument categories. The model, incorporating convolutional layers, batch normalization, and dropout regularization, achieved a peak validation accuracy of 78.37 % over 60 epochs. Comparative analysis with state-of-the-art methods highlights its competitive performance and computational efficiency. Robustness evaluations on varying input lengths and noise levels assess the model's generalization. Performance metrics, including accuracy trends, loss curves, and a confusion matrix, demonstrate strong classification for instruments like piano and violin while revealing challenges in distinguishing spectrally similar instruments. These findings underscore the effectiveness of CNNs for instrument classification and provide insights for enhancing deep learning-based audio recognition models.