Exploring accent bias in human and machine transcription

This project will explore how speaker accent affects the performance of both humans and machines when transcribing forensic recordings (i.e. those used in criminal investigations and court cases). Currently no research has investigated these system-level phenomena, despite transcription being a common task for forensic practitioners. With knowledge of how human/machine transcription behaviour is affected by accent, better, more impartial evidence can be presented in courts, and developers can make systems which use automatic speech recognition (like Siri and Alexa, or automated security systems) fairer to people of all accents.