Scholarlink

<p>Environmental sounds are crucial in understanding and monitoring diverse ecosystems, urban settings, and emergencies. Accurate classification of these sounds can enhance automated systems for surveillance, noise monitoring, and urban planning<b>.</b> This study investigates the application of Explainable Artificial Intelligence (XAI) techniques in urban sound recognition by integrating both Machine Learning (ML) and Deep Learning (DL) models. This research aims to enhance classification accuracy while ensuring transparency in model decision-making utilizing the UrbanSound8K (US8K) dataset, which consists of ten categories of urban sounds<b>.</b> The dataset comprises 8,732 audio samples covering various environmental and human-made sounds. We evaluate 11 Machine Learning models, including Bernoulli Naïve Bayes (BNB), AdaBoost, Quadratic Discriminant Analysis (QDA), Gaussian Naïve Bayes (GNB), Linear Discriminant Analysis (LDA), Support Vector Classifier (SVC), Multinomial Naïve Bayes (MNB), Logistic Regression (LR), Multilayer Perceptron (MLP), Nu-SVC, and K-Nearest Neighbors (KNN) and a DL model customized Convolutional Neural Network (CNN) model<b>.</b> Among these, our customized CNN model achieves the highest classification accuracy of 97.03 ± 0.2%. To ensure interpretability and trust in predictions<b>,</b> we integrate Local Interpretable Model-Agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP)<b>,</b> enabling insight into how specific features contribute to the model’s decision-making process. This research highlights the importance of XAI in sound classification, providing valuable insights into both model performance and prediction reasoning. Such advancements are crucial for developing reliable and transparent AI systems in real-world urban sound analysis.</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

Artificial intelligence-powered environmental sound recognition with explainable AI techniques

  • 期刊论文

  • 发表日期：

    2025年6月13日

查看全文
  • Farida Siddiqi Prity,
  • Md. Mubarak Hossain,
  • Md. Maruf Hossain,
  • Md. Shihab Uddin,
  • Md. Rabiul Islam,
  • Mirza Raquib,
  • Md. Rehan Ali,
  • K. M. Aslam Uddin

摘要

Environmental sounds are crucial in understanding and monitoring diverse ecosystems, urban settings, and emergencies. Accurate classification of these sounds can enhance automated systems for surveillance, noise monitoring, and urban planning. This study investigates the application of Explainable Artificial Intelligence (XAI) techniques in urban sound recognition by integrating both Machine Learning (ML) and Deep Learning (DL) models. This research aims to enhance classification accuracy while ensuring transparency in model decision-making utilizing the UrbanSound8K (US8K) dataset, which consists of ten categories of urban sounds. The dataset comprises 8,732 audio samples covering various environmental and human-made sounds. We evaluate 11 Machine Learning models, including Bernoulli Naïve Bayes (BNB), AdaBoost, Quadratic Discriminant Analysis (QDA), Gaussian Naïve Bayes (GNB), Linear Discriminant Analysis (LDA), Support Vector Classifier (SVC), Multinomial Naïve Bayes (MNB), Logistic Regression (LR), Multilayer Perceptron (MLP), Nu-SVC, and K-Nearest Neighbors (KNN) and a DL model customized Convolutional Neural Network (CNN) model. Among these, our customized CNN model achieves the highest classification accuracy of 97.03 ± 0.2%. To ensure interpretability and trust in predictions, we integrate Local Interpretable Model-Agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP), enabling insight into how specific features contribute to the model’s decision-making process. This research highlights the importance of XAI in sound classification, providing valuable insights into both model performance and prediction reasoning. Such advancements are crucial for developing reliable and transparent AI systems in real-world urban sound analysis.