Person Identification System from Speech and Laughter Using Machine Learning Algorithms
Oluwatoyin P. Popoola, Comfort Oluwaseyi Folorunso, Olumuyiwa Sunday Asaolu, J. J. Joshua, M. D. Oyeyemi · Journal of Engineering Research · 2020
Automated person identification and authentication is paramount for preclusion of cybercrime, national security and veracity of electoral processes. This is a critical component of Information and Communication Technology (ICT), which is the mainstay for national development. This paper presents the use of speech and laughter of people for person identification with the focus on forensics application where people speak and laugh in between. Features were extracted using the Librosa library in Python programming language via Scientific Python Development Environment (SPYDER) IDE (version 4.1.3) of the Anaconda software. While the Orange software (version 3.25.0) for data-mining was used for training, testing and validation of five standard machine learning algorithms: Neural Networks (NN), Support Vector Machine (SVM), Random Forest (RF), Naïve Bayes (NB) and Logistic Regression (LR). Results showed that the neural networks classifier gave the best accuracy followed by the SVM. There was an average of 17.6% and 14.1% increase in the validation metrics when both speech and laughter were combined as compared to speech and laughter independently respectively. This research area is very useful in forensics especially for recognising criminals in conversation.