Accent classification using Machine learning and Deep Learning Models

Accents are speech habits that different language speakers' exhibit. Usually, they have common traits for people of similar racial, cultural, or social context. Speech categorization is the question of finding accent type of the fellow citizens of different geographical locations. By speech classification, one can easily identify where the person belongs, his personal locations as well as religious backgrounds. But not always this speech classification thing is useful. Obviously, it is bound to make mistakes if done manually. Sometimes people refer to some accents or languages that hardly anyone uses or has ever heard of. This poses a challenge for most people as they do not understand it and fail to interpret the true meaning. Hence, this paper developed a model by utilizing machine learning and deep learning algorithms to create an effective tool for accent classification. The proposed system is validated using Speech Accent Archive dataset hosted by George Mason University on Kaggle. Results show that among various machine learning models, Decision Trees performed best even when it is compared to deep learning models such as CNN and LSTM and achieved an accuracy of 97.36%.