Daniel Tweddle

My name is Daniel, I’m a PhD researcher in speech recognition at the University of West London. My research area focuses on improving deep learning applications of mispronunciation detection and diagnosis (MDD), particularly for non-native speakers of a given dialect. For this to happen, we need two things: 1. Data - Native (L1) English speech with accurate pronunciation transcripts - Non-native (L2) English speech with accurate transcripts detailing the pronunciation differences relative to a defined L1, which we can label as 'errors' or 'mispronunciations'. 2. A deep learning system For British English pronunciation, there was a lack of suitable data, which led to a lack of deep learning model benchmarking, and a large research gap. To fill this, my work has included the construction of two novel speech datasets (MRP & MRPL2), and a novel pronunciation dictionary. These resources have allowed me to develop the first Transformer based MDD models for British English speech.