Sam O’Connor Russell, a PhD researcher at the ADAPT Centre and Trinity College Dublin, has had two first-author papers accepted at upcoming leading international conferences ACL and INTERSPEECH 2025. His work, supervised by Prof. Naomi Harte at the School of Engineering in TCD, explores predictive turn-taking models to help robots determine when to speak, which is a core challenge in human-robot interaction.
At ACL 2025, Sam introduced Multimodal-CAP, a GPT-based model that integrates both visual and acoustic signals. The work demonstrated that combining these cues significantly improves the model’s ability to predict speaker transitions in natural conversations.
At INTERSPEECH 2025, his second paper investigates the robustness of turn-taking models in noisy environments. Findings to be presented show that multimodal-VAP maintains a strong performance by relying on visual cues in the presence of noise which is a promising step toward real-word applications. In addition to his paper presentation, Sam will also share his PhD thesis plan at the INTERSPEECH Doctoral Consortium.
