L B S Institute of Technology for Women

The objective of emotional speech generation is to create synthetic speech that convincingly conveys specific emotions, enhancing the emotional quality of human–computer interactions. However, existing techniques often fall short of capturing the subtle emotional nuances, leading to speech that feels inauthentic. Additionally, many models lack the robustness needed to perform well across various emotional contexts, which limits their adaptability. Some methods may also generate overly exaggerated or artificial emotional responses, diminishing their effectiveness in real-world scenarios. This research explores using Generative Adversarial Networks (GAN) combined with Convolutional Neural Networks (CNN) for emotional speech generation. The process begins with audio preprocessing using Mel spectrograms for noise reduction and min–max normalization. A CNN-based GAN is then applied for feature extraction. The combination of CNN and GAN is used to classify emotions such as fear, anger, sadness, and happiness from the extracted features. The performance of the proposed method was evaluated using two datasets: RAVDESS and IEMOCAP. Results show that this approach can effectively detect speech emotions, achieving average accuracies of 99

With the rise of complex digital infrastructure in the modern era and the multifaceted nature of threats moving across networks, no single intrusion detection system can effectively defend against them. Any of the tools available today are making false positive rates seem like a problem that cannot be solved because they are only tested against a limited subset of the attack scenarios they claim to detect, or they are not tested against dynamic attacks. Not only can they respond in real time and have low false alarms (readily accepted as legit), but these techniques also generalize poorly due to their dependence on humans to step in for gaps in knowledge about threat capabilities, and so are similarly limited to traditional human-based systems. To address these problems, this work introduces SentinelAI-IDS. This original deep learning-based hybrid intrusion detection architecture leverages enhanced Bayesian optimization to optimize a random forest and XGBoost ensemble classifier, thereby addressing the robustness and generalization problems in IDS research. SentinelAI-IDS is capable of multi-dataset training on CICIDS_2017, UNSW_NB15, and NSL-KDD; thus, it is intended to offer a broad defense approach across datasets. We show that MI-based feature selection + PCA is a strong learning pipeline and can significantly reduce dimensionality. The hybrid model accumulates and then qualitatively analyzes a set of traffic data streams using simple majority voting for efficient classification and automated responses. Currently, SentinelAI-IDS surpasses all existing state-of-the-art methods available in the literature, achieving 97.10

In the current century, most human beings are distressed by stress. The stress impacts the mental and physical health of people and causes a high influence on human health. The stress affects the daily activities of human life like academics and work. Moreover, stress causes illnesses like anxiety, headaches, heart disease, and depression. Hence, it is essential to avoid these kinds of negative issues due to the high level of stress. The earlier stage of stress detection is necessary to reduce the stress level and provide proper treatment to patients. In this paper, the hybrid deep learning (DL) model called cascaded neuro-fuzzy SpinalNet (CNFSNet)–based human stress level detection model is proposed. The min–max normalization–enabled data normalization is the initial process, in which the data is normalized. Moreover, the essential features from the data are augmented to enhance the data sizes. The fuzzy local information cluster means (FLICM) is employed for feature clustering. At last, the detection of stress levels is performed using the CNFSNet. The accuracy, precision, and recall metrics are used to estimate the CNFSNet-based stress level detection model, with the outcomes of 0.9012, 0.906, and 0.902 achieved.

Cardiomegaly, or an enlarged heart, is a key indicator of various cardiac diseases, including Atrial Septal Defect, and an early and accurate diagnosis is crucial for timely medical intervention. However, current automated systems for cardiomegaly detection face challenges such as limited generalization, low accuracy in complex cases, and difficulty in segmenting relevant regions from chest X-rays. Recent advancements in deep learning, such as the use of ResNet50 for feature extraction and TernausNet for segmentation, have shown promise, yet these techniques often struggle with feature fusion and regional focus in medical images. In this paper an advanced deep learning architecture, InsTerNet is proposed that integrates ResNet50 for robust feature extraction with TernausNet for enhanced segmentation, specifically addressing these challenges. The performance of InsTerNet is evaluated using the reputed chest radiograph dataset obtained from the National Institutes of Health (NIH), demonstrating it to be better with AUROC of 0.97. The training accuracy obtained is 0.99 with the test accuracy of 0.94. The value of precision obtained is 0.92 with the recall and F1 score values 0.96, and 0.94, respectively, which is better compared to state-of-the-art techniques, making it a possible automated diagnostic tool to detect cardiomegaly.