This repository deals with Deception Detection using Audio Classification The main objective of this project is to use the verbal/speech modalities to detect deception adopted by humans. The implementation of the deep learning methods will surely outperform the human capabilities to identify deceit.
The most innovative part of our project is the type of data used i.e. audio/speech. If we use video data then it can lead to racial profiling and numerous other physical biases which will give us numerous false positives and negatives. It also makes collection of data easier and more uniform and is able to include and adapt to more outliers. Moreover, in case of video data, too many different dependencies like angle, lighting, etc. come into picture which can skew the data.
The modules in the project are as follows: • IMPORTING DATA • CONVERTING TO SPECTOGRAM • CONVERTING TO MFCC • CLEANING AND AUGMENTATION OF DATA • BUILDING NETWORK ARCHITECTURE • TRAINING NETWORK ARCHITECTURE ON DATA • DECIDING EVALUATION METRICS AND TESTING
Dataset used: https://web.eecs.umich.edu/~mihalcea/downloads.html#RealLifeDeception
Contributors : @Rixhi7 , @Achinthya2411 , @ayushkr03