Classification of Sound using Convolutional Neural Networks

• Published in: 2022 5th International Conference on Contemporary Computing and Informatics (IC3I) pp. 1015 - 1019
• Main Authors: Chaturvedi, Abhay, Yadav, Suman Avdhesh, Salman, Hayder M, Goyal, Himanshu Rai, Gebregziabher, Haftom, Rao, A Kakoli
• Format: Conference Proceeding
• Language: English
• Published: IEEE 14.12.2022
• Subjects: Convolutional Neural Network; Sound Classification; Spectrogram Images; Speech recognition
• DOI: 10.1109/IC3I56241.2022.10072823

Sound has a significant impact on every aspect of human life. The study of sound classification has gained popularity recently across a wide range of fields. In a variety of industries, from essential surveillance to personal security, sound is frequently used to develop automated systems. Speech recognition is becoming increasingly crucial as technology advances, and artificial intelligence is a key component. The sound classifier systems are developed that address the efficiency issues of conventional systems by utilizing deep learning architectures. In this paper, it offers a deep learning method for classifying environmental noise based on generated spectrograms. The sounds that the people hear most frequently can now be detected by deep learning systems. Convolutional neural networks have recently been used to classify environmental noise. Using CNN spectrogram images, a convolutional neural network is trained to detect environmental noise. The performance recognition has been greatly enhanced by models of sound categorization systems based on convolutional neural networks (CNN). Large datasets with a greater number of labels are often needed for high-performance classification. The audio was processed using Mel spectrogram model 1, Mel spectrogram model 2 and MFCC spectrogram model and then transformed into an image. Based on the results, it is concluded that CNN model achieves 96.7% accuracy by utilizing hybrid spectrogram model to build a noise classification system.