Scholarlink

<p>Bangla Speech Emotion Recognition (SER) is crucial for systems that respond to human emotions, such as in healthcare, education, and safety. However, because to poor datasets, overfitted models, and a dearth of research on adaptive learning and real-world deployment, the discipline has not made satisfactory progress. Most prior research trains on a single small dataset, uses overly complex models, and rarely tests edge feasibility, resulting a gap between research accuracy and practical use. Our work addresses this through four experiments: comparing single versus merged dataset training, large handcrafted versus MFCC-only feature sets, the effect of noise on robustness, and architectures for reducing overfitting while maintaining efficiency. The final system uses MFCC features with a CNN–BiLSTM model, balancing local spectral and temporal patterns, with dropout and careful validation to prevent overfitting. It reaches 82% accuracy across seven Bangla emotions, surpassing prior baselines, while running smoothly and efficiently on a Raspberry Pi 4B (~2.83 W). Guided by affective computing, the system couples emotion detection with appropriate and context‑sensitive responses rather than treating recognition as an end point. Evidence from emotion‑sensitive tutoring and the cognitive‑disequilibrium theory motivates the use of SER to sustain productive confusion, prevent frustration–boredom cycles, and support deeper learning through adaptive dialogue. Accordingly, SER signals guide a language model to adjust tone, pacing, and scaffolding while preserving learner agency via Human–AI shared regulation and calibrated human involvement in both supervisory and assistive roles. An incremental learning module enables continual updates without requiring complete retraining for long-term personalization, and SER outputs direct a language model to provide emotionally adaptive answers in addition to static classification. The first Bangla SER–LLM integration with an incremental learning pathway for sustained adaptation is presented, along with examples of how multi-dataset integration and noise handling enhance generalization, demonstrate that MFCC features alone can compete with larger sets, achieve state-of-the-art Bangla SER under low-resource conditions, and validate real-time Raspberry Pi deployment.</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

Emotionally Aware Bangla Speech Systems: Real-Time SER with Adaptive Learning and LLM Integration

  • 期刊论文

  • 发表日期：

    2026年2月2日

查看全文
  • Mostakim Hossain,
  • Md. Sakibul Alam Patwary,
  • Md. Musfiq Hossain,
  • Rashedur M. Rahman

摘要

Bangla Speech Emotion Recognition (SER) is crucial for systems that respond to human emotions, such as in healthcare, education, and safety. However, because to poor datasets, overfitted models, and a dearth of research on adaptive learning and real-world deployment, the discipline has not made satisfactory progress. Most prior research trains on a single small dataset, uses overly complex models, and rarely tests edge feasibility, resulting a gap between research accuracy and practical use. Our work addresses this through four experiments: comparing single versus merged dataset training, large handcrafted versus MFCC-only feature sets, the effect of noise on robustness, and architectures for reducing overfitting while maintaining efficiency. The final system uses MFCC features with a CNN–BiLSTM model, balancing local spectral and temporal patterns, with dropout and careful validation to prevent overfitting. It reaches 82% accuracy across seven Bangla emotions, surpassing prior baselines, while running smoothly and efficiently on a Raspberry Pi 4B (~2.83 W). Guided by affective computing, the system couples emotion detection with appropriate and context‑sensitive responses rather than treating recognition as an end point. Evidence from emotion‑sensitive tutoring and the cognitive‑disequilibrium theory motivates the use of SER to sustain productive confusion, prevent frustration–boredom cycles, and support deeper learning through adaptive dialogue. Accordingly, SER signals guide a language model to adjust tone, pacing, and scaffolding while preserving learner agency via Human–AI shared regulation and calibrated human involvement in both supervisory and assistive roles. An incremental learning module enables continual updates without requiring complete retraining for long-term personalization, and SER outputs direct a language model to provide emotionally adaptive answers in addition to static classification. The first Bangla SER–LLM integration with an incremental learning pathway for sustained adaptation is presented, along with examples of how multi-dataset integration and noise handling enhance generalization, demonstrate that MFCC features alone can compete with larger sets, achieve state-of-the-art Bangla SER under low-resource conditions, and validate real-time Raspberry Pi deployment.