While previous gesture generation studies have highlighted the benefits of deep learning-based approaches for generating human-like gestures, these often require large datasets and intensive computation. Our model differentiates between short and long dialogues, generating context-specific gestures for short exchanges (e.g., greetings, farewells, agreement/disagreement) and emotion-based gestures for longer dialogues (neutral, happy, aggressive). We compare the system’s performance against ground truth gestures, random gestures, and idling gestures using metrics from the GENEA Challenge. This approach aims to provide a more efficient alternative to deep learning models. Our findings are expected to contribute to the development of more engaging, responsive virtual assistants, improving user comprehension in human-computer interaction.
Read the full publication: https://doi.org/10.26760/elkomika.v12i4.953
