Scholarlink

<p>Neural decoding of speech intention could advance the development and application of brain-computer interface (BCI) technology. Currently, lack of dataset limited the research on decoding the true speech intention, especially the diverse intentions expressed by the same text when no context is given. This study provides an EEG dataset, CIRE, on spoken language interaction intention featuring aligned textual expressions with divergent intentional meanings due to the differences in prosodic emotion. The dataset comprises preprocessed high-density (128-channel) EEG recordings from 38 participants engaged in comprehension of attitude-conveying speech stimuli, accompanied by Wav2vec2-derived acoustic embeddings of the listening materials. To validate our dataset through cognitive neuroscience studies and binary intent classification, we applied signal processing pipelines, cognitive analysis frameworks, and machine learning (ML) approaches. Our baseline model achieved a cross-subject classification accuracy of 68.2%, with differences exhibiting interpretable neurophysiological correlates. The high-density and high temporal resolution EEG data offer broader application areas, both in cognitive neuroscience and speech BCI, and can also contribute to the brain-inspired algorithms.</p>

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

CIRE: A Chinese EEG Dataset for decoding speech intention modulated by prosodic emotion

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月21日

查看全文
  • Shengrui He,
  • Zhongjie Li,
  • Jianwu Dang,
  • Yingyi Luo,
  • Gaoyan Zhang

摘要

Neural decoding of speech intention could advance the development and application of brain-computer interface (BCI) technology. Currently, lack of dataset limited the research on decoding the true speech intention, especially the diverse intentions expressed by the same text when no context is given. This study provides an EEG dataset, CIRE, on spoken language interaction intention featuring aligned textual expressions with divergent intentional meanings due to the differences in prosodic emotion. The dataset comprises preprocessed high-density (128-channel) EEG recordings from 38 participants engaged in comprehension of attitude-conveying speech stimuli, accompanied by Wav2vec2-derived acoustic embeddings of the listening materials. To validate our dataset through cognitive neuroscience studies and binary intent classification, we applied signal processing pipelines, cognitive analysis frameworks, and machine learning (ML) approaches. Our baseline model achieved a cross-subject classification accuracy of 68.2%, with differences exhibiting interpretable neurophysiological correlates. The high-density and high temporal resolution EEG data offer broader application areas, both in cognitive neuroscience and speech BCI, and can also contribute to the brain-inspired algorithms.