Scholarlink

Lack of communication increases health risks such as depression and dementia, making conversational partners essential for elderly support. Conversational agents or robots have the potential to assist in this area. To achieve natural human-agent listening interactions, appropriate responses to the human speaker’s level of engagement are essential. This study investigates the effectiveness of integrating recently emerging LLM-based features into engagement estimation. Three machine learning experiments are conducted to compare various architectural designs using non-verbal video and audio features (facial expressions, gaze, head movements, and prosody), partial or full utterances, and time-series or flattened fixed-size vectors.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Investigation into the Use of LLM-Based Features for Multimodal Estimation of Elderly Engagement in Active Listening

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hung-Hsuan Huang,
  • Koichi Takahashi

摘要

Lack of communication increases health risks such as depression and dementia, making conversational partners essential for elderly support. Conversational agents or robots have the potential to assist in this area. To achieve natural human-agent listening interactions, appropriate responses to the human speaker’s level of engagement are essential. This study investigates the effectiveness of integrating recently emerging LLM-based features into engagement estimation. Three machine learning experiments are conducted to compare various architectural designs using non-verbal video and audio features (facial expressions, gaze, head movements, and prosody), partial or full utterances, and time-series or flattened fixed-size vectors.