The GENEA Challenge 2026 evaluated five speech-driven gesture generation systems using the Seamless Interaction dataset. The challenge employed a disentangled methodology to assess motion quality and speech alignment separately, alongside a dyadic mismatching study to understand how systems react to interlocutors. New tasks included semantic gesture generation and text-mismatching evaluation. Results showed that while the dataset's motion-capture segments significantly outperformed all submissions in realism and speech alignment, the challenge systems struggled to generate semantically expressive gestures or respond appropriately to conversational partners. AI
IMPACT AI systems for generating speech-driven gestures still significantly lag behind human capabilities in realism, speech alignment, and semantic expressiveness.
RANK_REASON The item is an academic paper detailing the results of a challenge evaluating AI systems. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
