Scholarlink

<p>Speech is the important form of human communication which is also utilized while interaction with computers. Speech emotional recognition is a way of determining human emotions from speech. With applications in fields like computational neuroscience, cognitive psychology, intelligent tutoring, and healthcare, affective computing and human-machine interaction are growing fields of study. Although previous research indicates that accents may affect how people express their emotions but their impact on SER has not been systematically investigated. We created a new accented Urdu corpus called Accented-SEMOUR with 15,040 audio clips recorded by eight performers expressing eight main emotions to bridge this gap. We carried out two experiments using deep neural network models to investigate how accents affect the accuracy of SER in the Urdu language. Speech accents were added to the Urdu language in the first experiment, and accent diversity was added in the second. Our findings show that augmenting the original Urdu dataset with accented speech significantly enhanced recognition performance, attained 97.77% training accuracy and 95.93% validation accuracy. These results indicate that accent information significantly improves Urdu SER and deliver the first empirical findings showing the influence of accent on emotion recognition. This work presents the first accented Urdu SER corpus and provides novel insights to advance SER in low-resource and multilingual contexts.</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

Impact of accent on urdu speech emotion recognition: Deep learning based an experimental study

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Sadaqat Ali,
  • Hajra Murtaza,
  • Musharraf Ahmed

摘要

Speech is the important form of human communication which is also utilized while interaction with computers. Speech emotional recognition is a way of determining human emotions from speech. With applications in fields like computational neuroscience, cognitive psychology, intelligent tutoring, and healthcare, affective computing and human-machine interaction are growing fields of study. Although previous research indicates that accents may affect how people express their emotions but their impact on SER has not been systematically investigated. We created a new accented Urdu corpus called Accented-SEMOUR with 15,040 audio clips recorded by eight performers expressing eight main emotions to bridge this gap. We carried out two experiments using deep neural network models to investigate how accents affect the accuracy of SER in the Urdu language. Speech accents were added to the Urdu language in the first experiment, and accent diversity was added in the second. Our findings show that augmenting the original Urdu dataset with accented speech significantly enhanced recognition performance, attained 97.77% training accuracy and 95.93% validation accuracy. These results indicate that accent information significantly improves Urdu SER and deliver the first empirical findings showing the influence of accent on emotion recognition. This work presents the first accented Urdu SER corpus and provides novel insights to advance SER in low-resource and multilingual contexts.