Speech-based identification of social groups in a single accent of British English by humans and computers

Abstract

Classification of social groups within a given accent is a challenging refinement of language identification (LID) and accent/dialect recognition. The 2001 census of England and Wales identifies two main ethnic groups in the city of Birmingham, which it refers to as Asian and white. In this paper LID techniques are applied to the problem of identifying individuals from these two groups who were born in Birmingham and hence speak British English with a Birmingham accent. An Equal Error Rate (EER) of 3.57% is obtained using a LID system which fuses the outputs of several acoustic and phonotactic systems. This performance is much better than expected and compares to an EER of 8.72% achieved by human listeners. The implications of this result for automatic speech recognition are discussed