This study explores lie detection by focusing on sound. By using real-world courtroom data changed into sound, a Convolutional Neural Network (CNNConvolutional Neural Network (CNN)) model is used to find important features from sound data that help in spotting lies. Unlike traditional methods that need manual work to identify features, the CNNCNN (ConvNet) model does this work on its own and shows a high accuracy of 93.5%. This improvement is not only good for finding lies in sound but also helps in improving the performance of systems that use multiple types of data for lie detection. This makes it a strong and reliable tool for real-life courtroom scenarios. The results show that CNNCNN (ConvNet) is very good at handling the complex structures of sound data, moving us closer to better lie detection methods. Moreover, the study points out the need for more real and diverse data to help further improve lie detection. This will help in enhancing trust and safety in different areas.