Classification on distressed sounds with CNN/RNN

• Main Author: Guo, Xihuang
• Other Authors: Gan Woon Seng
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/139181
• Institution: Nanyang Technological University

Nowadays, people pay more attention to their personal safety due to the improvements in their quality of life. Imagine if you called for a policeman for help, they would be able to arrive within minutes and that could reduce the chance of crime. This can be done by classifying the distressed sounds using Machine Learning. This project can be integrated with a sound-based security system to help those people who need urgent help or assistance. In this report, it focuses on how to classify a distressed sound using the Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN). In particular, the dataset was collected for 3 distressed sounds, “Help”, “Crying” and “Screaming”, and then built a model to determine which distressed sound among them. The model to be implemented is a VGG which is widely used in audio classification. The report shows how to convert an audio classification problem to image recognition, where the fully developed techniques of CNN and RNN can be applied better. In the end, the performance of these two networks was evaluated based on several properties. CNN model performs better in overall with 94% training accuracy and 85% testing accuracy.