The immense progress in physiological signal acquisition and processing in health monitoring allowed a better understanding of patient disease detection and diagnosis. With the increase in data volume and power consumption, effective data compression, signal acquisition, transmission, and processing techniques are essential, especially in telemonitoring healthcare applications. An emerging research area focuses on integrating Compressed Sensing (CS) with physiological signals to deal with a massive amount of physiological data, transmission bandwidth, and power-saving purposes. A review of compressed sensing for physiological signals is presented in this paper, including electroencephalography (EEG), electrocardiography (EEG), electromyography (EMG), and electrodermal activity (EDA), focusing on the pros and cons of CS in treating such signals and the suitability of CS for hardware implementation. Furthermore, we emphasize performance matrices such as Compression Ratio (CR), Signal to-Noise-Ratio (SNR), Percentage Root-mean-square Difference (PRD), and Processing time to evaluate the performance of CS. We also investigate the current practices, challenges, and opportunities of using CS in healthcare applications.

Speech Emotion Recognition (SER) technology helps computers understand human emotions in speech, which fills a critical niche in advancing human–computer interaction and mental health diagnostics. The primary objective of this study is to enhance SER accuracy and generalization through innovative deep learning models. Despite its importance in various fields like human–computer interaction and mental health diagnosis, accurately identifying emotions from speech can be challenging due to differences in speakers, accents, and background noise. The work proposes two innovative deep learning models to improve SER accuracy: a CNN-LSTM model and an Attention-Enhanced CNN-LSTM model. These models were tested on the Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS), collected between 2015 and 2018, which comprises 1440 audio files of male and female actors expressing eight emotions. Both models achieved impressive accuracy rates of over 96% in classifying emotions into eight categories. By comparing the CNN-LSTM and Attention-Enhanced CNN-LSTM models, this study offers comparative insights into modeling techniques, contributes to the development of more effective emotion recognition systems, and offers practical implications for real-time applications in healthcare and customer service.

This paper introduces a novel Compressive Sensing (CS)-based cryptosystem tailored for the secure and efficient transmission of Electrocardiogram (ECG) signals in Internet of Medical Things (IoMT) environments. It leverages the inherent sparsity of ECG signals in the wavelet domain and ensures both data privacy and integrity during transmission through a simple but effective additional encryption stage. We have evaluated the performance of four distinct sensing matrices and three reconstruction algorithms across multiple wavelet families. Through extensive simulations using the MIT-BIH Arrhythmia Database we demonstrate that the Low-Density Parity-Check matrix combined with the L1 optimization algorithm achieves the highest Quality Score, with Compression Ratio up 50%. The proposed approach also shows an excellent ability in preserving important pathological features in presence of abnormal beats. The proposed CS encoder has been hardware implemented on low-resource microcontroller and FPGA devices. When realized on a Xilinx Artix 7 XC7A12 T FPGA, such a prototype allows real-time operations to be sustained running at 1 MHz clock frequency and dissipating only 0.8nJ per sample.

Electrocardiograms (ECGs) are essential for diagnosing cardiac conditions, yet high-fidelity 12-lead signals pose challenges for storage, transmission, and real-time processing. Compressed Sensing (CS) addresses these issues by exploiting signal sparsity, but its reliance on complex reconstruction algorithms limits practical deployment. Compressed Learning (CL) offers a more efficient alternative by enabling direct feature extraction from compressed measurements. This paper proposes an optimized CL framework for classifying arrhythmia (ARR), congestive heart failure (CHF), and normal sinus rhythm (NSR) from compressed ECG signals. We introduce a deterministic block-sparse binary sensing matrix with diagonal block allocation of ones and zeros elsewhere. This lightweight design eliminates floating-point multiplications, reduces memory overhead, and ensures uniform computational complexity across compression ratios. Compared with widely used Gaussian, Bernoulli, and Fourier sensing matrices, the proposed approach preserves ECG morphology more effectively and is well suited for embedded hardware. A deep learning-based Convolutional Neural Network (CNN) is developed and benchmarked against conventional machine learning classifiers, demonstrating superior classification accuracy across all tested compression ratios redachieving up to 96.56% accuracy, 96.5% F1-score, and 97.9% specificity at CR = 0.5, and outperforming Gaussian and Bernoulli matrices. To validate real-time feasibility, the proposed matrix is implemented on an STM32 NUCLEO-F401RE microcontroller. Hardware experiments confirm that the framework achieves high accuracy while significantly reducing energy consumption and computational load requiring only 9 μs per compression operation and consuming 0.641 μJ with constant-time execution across CRs. Furthermore, the proposed framework is validated on three widely used PhysioNet databases (MIT-BIH Arrhythmia, Normal Sinus Rhythm, and BIDMC Congestive Heart Failure), which together capture diverse patient populations and varying acquisition devices. This ensures robustness of the method and supporting its applicability in diverse clinical environments. Extending this framework to wearable ECG systems represents a promising direction for future research.