Identification of Sounds Using Deep Learning with MFCC Features Extraction

Environmental sound classification is an essential aspect of identifying different classes of sounds using a machine learning approach. This work investigates the efficacy of convolutional neural networks for classifying short environmental sound clips. In this work, a deep 1D-CNN architecture with 2 layers of convolutional employing max-pooling, followed by two fully-connected layers are utilized. The model leverages segmented spectrograms with delta features as input. Evaluated with datasets encompassing environmental and urban recordings, our model surpasses baseline methods utilizing mel-frequency cepstral coefficients (MFCCs) and achieves competitive performance against traditional approaches.