Enhancing Environmental Sounds Classification through Deep Learning Techniques

Environmental sound classification has become an important application for the research process in recent years, as it has many applications in several fields such as urban noise management, wildlife monitoring, and intelligent sound system. Sound classification involves mainly classifying the different sounds and predicting the class of that sound by using deep learning techniques. These techniques have proven to be very effective in many sound analysis and classification areas. This study uses neural networks to learn high-level features from audio clips, then build some layers, and finally make a fully connected layer to know the final classification. Generally, Mel-Frequency Cepstral Coefficients (MFCCs) have been taken as a feature representation for audio signals. These features can be generated from the spectrograms, which are inputs to the organized model. The model can be built using networks like artificial neural networks and convolution neural networks. The architecture of the neural network is designed to effectively capture similar features of environmental sounds by reducing the competition of complexity. To implement the sound classification, a UrbanSoundDataset8k is used to predict the individual classes.