A Performance Study: Convolutional Deep Belief Networks and Convolutional Neural Networks for Audio Classification

This paper presents an innovative approach for supervised audio classification using Convolutional Deep Belief Networks (CDBN). Using advanced Digital Signal Processing techniques, the method efficiently extracts salient features from audio signals, enabling the training of a robust Convolutional Deep Belief Network (CDBN). The performance of the proposed method is then evaluated and compared with that of Convolutional Neural Network(CNNs)-based audio classifiers trained on the same labeled data. The paper rigorously evaluates the pro-posed method against annotated data and different MFCC tuning parameters, showing its performance on par with advanced CNN-based audio classifiers. A remarkable accuracy of 94.05 % was achieved for the DBN model compared to 94.05% by the CNN model. The effect of changing various MFCC parameters on the model performance is also shown. Additionally, the paper analyzes the pros and cons of Deep Belief Networks for audio classification, offering valuable insights for future research in this field.