[Google Scholar] [Github] [CV] [LinkedIn] [Twitter]
Matthew Roddy is a PhD candidate at the Electrical Engineering Department of Trinity College Dublin. His research focuses on the design of conversational turn-taking models that can aid spoken dialogue systems (SDSs) in producing naturalistic interactions. More specifically, his work focuses on drawing information from multiple modalities to train deep learning models to predict conversational turn-taking behavior. His research incorporates elements from machine-learning, audio-visual signal processing, natural language processing, and social sciences.
His supervisor is Dr. Naomi Harte.
He is part of the Sigmedia Lab and is funded by the ADAPT Centre.
Expected thesis submission date: Summer 2019.
- Roddy, Matthew, Gabriel Skantze, and Naomi Harte. “Multimodal Continuous Turn-Taking Prediction Using Multiscale RNNs.” ICMI, 2018. [ accepted ] [ pdf ] [ code ]
- Roddy, Matthew, Gabriel Skantze, and Naomi Harte. “Investigating Speech Features for Continuous Turn-Taking Prediction Using LSTMs.” INTERSPEECH, 2018. [ accepted ] [ pdf ] [ code ]
- Roddy, Matthew, and Naomi Harte. “Detecting conversational gaze aversion using unsupervised learning.” European Signal Processing Conference (EUSIPCO), 2017. [ pdf ]
- Roddy, Matthew, and Naomi Harte. “Towards predicting dialog acts from previous speakers’ non-verbal cues.” Symposium on Multimodal Communication (MMSYM), 2017. [ pdf ]
- Roddy, Matthew, and Jacqueline Walker. “A Method of Morphing Spectral Envelopes of the Singing Voice for Use with Backing Vocals.” DAFx, 2014. [ pdf ]