Classification of English Accents

Accent is a strong form of language variation that introduces many challenges in today’s 
multicultural societies starting from meeting rooms to automated call centers. In this 
project we aim at a data driven approach for classifying English accents. To start with, 
we investigate several classification techniques such as support vector machine, decision 
tree and k-nearest neighbors to classify three English accents; Spanish, Mandarin and 
Arabic on a data collected from the GMU speech accent archive. We achieve 65% accuracy 
with standard RASTA-PLP features. Our future plan includes segmenting the speech samples 
in phonemes to perform classification at the phoneme level and collecting more labeled data 
from the crowd via game app.