Scholarlink

This paper develops a durable Speech Emotion Recognition system using Domain Adversarial Neural Networks that uses Hindi language samples. Our work uses transfer learning techniques to address the primary problem of limited labeled emotional data for Hindi. Our target domain has a unique Hindi audio dataset, whereas a source domain uses the CREMA-D dataset. Emotions present in both datasets are retained during the domain compatibility selection process. For Hindi and English audio domain shift reduction, the model uses DANN, which increases the efficacy of generalization. Despite showing significant superiority over traditional techniques, the developed DANN model achieves an accuracy level of 91% on the Hindi dataset.The study advances multilingual SER systems with a particular focus on Hindi and has potential uses in affective computing and human-computer interaction.

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

Cross-Corpus Speech Emotion Recognition via Domain-Adversarial Neural Network

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • G. V. N. S. Harsha Vardhan,
  • Shaik Afreen Sultana,
  • Shaziya Mahi,
  • Syed Shareefunnisa

摘要

This paper develops a durable Speech Emotion Recognition system using Domain Adversarial Neural Networks that uses Hindi language samples. Our work uses transfer learning techniques to address the primary problem of limited labeled emotional data for Hindi. Our target domain has a unique Hindi audio dataset, whereas a source domain uses the CREMA-D dataset. Emotions present in both datasets are retained during the domain compatibility selection process. For Hindi and English audio domain shift reduction, the model uses DANN, which increases the efficacy of generalization. Despite showing significant superiority over traditional techniques, the developed DANN model achieves an accuracy level of 91% on the Hindi dataset.The study advances multilingual SER systems with a particular focus on Hindi and has potential uses in affective computing and human-computer interaction.