Scholarlink

<p>Neonatal cry classification using audio signals has emerged as a promising non-invasive approach for the early identification of health-related abnormalities in newborns. Since infant cries reflect underlying physiological and neurological conditions, systematic analysis of cry acoustics can support timely clinical intervention and improve neonatal outcomes. However, existing cry-based classification systems often suffer from limited robustness due to background noise, cry pattern variability, and insufficient feature representation. To address these challenges, this paper proposes an advanced acoustic signal processing framework termed Cry-P3CNN for accurate neonatal cry classification. Initially, raw cry audio signals undergo enhanced preprocessing, where noise components are suppressed using Enhanced Variational Mode Decomposition (VMD), followed by Min–Max normalization to standardize signal amplitudes. Subsequently, novel Parallel Three-Tier Convolutional Neural Network (P3CNN) architecture is employed to extract time-invariant and discriminative features through three complementary pathways: Flexible Feature Enhancer (F2E), Pattern Extractor (PE), and Frequency Extractor (FE). To further reduce redundancy and improve feature discrimination, a lightweight Shamble Attention Mechanism (SAM) is integrated to adaptively emphasize informative feature representations. Finally, neonatal cry classification is performed using an Improved Reinforcement Learning (IRL) strategy based on Double Deep Q-Learning, enabling adaptive decision-making and enhanced generalization capability. The proposed Cry-P3CNN model is evaluated on publicly available neonatal cry datasets and compared with several state-of-the-art methods using standard performance metrics, including accuracy, precision, recall, and F1-score. Experimental results demonstrate that Cry-P3CNN consistently outperforms existing approaches, confirming its effectiveness and reliability. Overall, this study highlights the potential of combining advanced signal preprocessing, parallel deep feature extraction, attention mechanisms, and reinforcement learning to develop a robust and scalable neonatal cry classification system for early health assessment.</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

Cry-P3CNN: audio signal processing based neonatal cry classification using improved reinforcement learning

  • 期刊论文

  • 发表日期：

    2026年7月11日

查看全文
  • Kumar Perumal,
  • Shakir Khan,
  • Rakesh Kumar Mahendran,
  • Thamer Alshammari

摘要

Neonatal cry classification using audio signals has emerged as a promising non-invasive approach for the early identification of health-related abnormalities in newborns. Since infant cries reflect underlying physiological and neurological conditions, systematic analysis of cry acoustics can support timely clinical intervention and improve neonatal outcomes. However, existing cry-based classification systems often suffer from limited robustness due to background noise, cry pattern variability, and insufficient feature representation. To address these challenges, this paper proposes an advanced acoustic signal processing framework termed Cry-P3CNN for accurate neonatal cry classification. Initially, raw cry audio signals undergo enhanced preprocessing, where noise components are suppressed using Enhanced Variational Mode Decomposition (VMD), followed by Min–Max normalization to standardize signal amplitudes. Subsequently, novel Parallel Three-Tier Convolutional Neural Network (P3CNN) architecture is employed to extract time-invariant and discriminative features through three complementary pathways: Flexible Feature Enhancer (F2E), Pattern Extractor (PE), and Frequency Extractor (FE). To further reduce redundancy and improve feature discrimination, a lightweight Shamble Attention Mechanism (SAM) is integrated to adaptively emphasize informative feature representations. Finally, neonatal cry classification is performed using an Improved Reinforcement Learning (IRL) strategy based on Double Deep Q-Learning, enabling adaptive decision-making and enhanced generalization capability. The proposed Cry-P3CNN model is evaluated on publicly available neonatal cry datasets and compared with several state-of-the-art methods using standard performance metrics, including accuracy, precision, recall, and F1-score. Experimental results demonstrate that Cry-P3CNN consistently outperforms existing approaches, confirming its effectiveness and reliability. Overall, this study highlights the potential of combining advanced signal preprocessing, parallel deep feature extraction, attention mechanisms, and reinforcement learning to develop a robust and scalable neonatal cry classification system for early health assessment.