This research was conducted by a team of three undergraduate Computer Science students at Wentworth Institute of Technology in July 2020 to explore the possible applications of Machine Learning to the classification of real-world voice data. The focus of the work was to experiment with various methods of feature engineering and data analysis, along with the performance of various Machine Learning models in this application. The team focused on applying tools available in the scikit-learn Python library to apply common Machine Learning algorithms to the problem. The employed methods had the ability to demonstrate the effectiveness of both feature engineering and data processing, as well as various Machine Learning models. The results indicate that using simple machine learning models, audio speaker classification is possible. After data tuning, model selection, and hyperparameter tuning, we were able to achieve an accuracy of 74% between 7 classes. This level of result using only traditional machine learning models shows that this approach with a more modern deep learning algorithm could provide much more accurate and useful results.

[1] A. Nagrani, J. S. Chung, W. Xie and A. Zisserman, "Voxceleb: Large-scale speaker verification in the wild," elsevier, p. 15, 2019.

[2] J. Hrisko, "Audio Processing in Python Part I: Sampling, Nyquist, and the Fast Fourier Transform," 13 September 2018. [Online].