English accents classification using Deep Learning

Abstract

Existing voice recognition systems achieve solid results in processing human speech, however there are still a lot of factors that strongly affect their performance. Accents vary across the world and strongly affect people’s speech thus that may cause significant problems for a computer to process the speech. In some cases, there is no clear possibility to develop a system that would be equally effective for all the accents. For the sake of addressing this issue, various techniques are used to mitigate this problem. This project is aimed at investigating the performance of different deep learning models such as a Multi-Layer Perceptron and a Convolutional Neural Network for the task of classifying several English accents and to suggest a model that would achieve the best performance on the Speech Archive dataset.