Those with disabilities who have lost their legs must use a prosthesis to walk. However, traditional prostheses have the disadvantage of being unable to move and support the human gait because there are no mechanisms or algorithms to control them. This makes it difficult for the wearer to walk. To overcome this problem, we developed an insole device with a wearable sensor for real-time gait phase detection based on the kNN (k-nearest neighbor) algorithm for prosthetic control. The kNN algorithm is used with the raw data obtained from the pressure sensors in the insole to predict seven walking phases, i.e., stand, heel strike, foot flat, midstance, heel off, toe-off, and swing. As a result, the predictive decision in each gait cycle to control the ankle movement of the transtibial prosthesis improves with each walk. The results in this study can provide 81.43% accuracy for gait phase detection, and can control the transtibial prosthetic effectively at the maximum walking speed of 6 km/h. Moreover, this insole device is small, lightweight and unaffected by the physical factors of the wearer.

Recent studies have shown the potential of the Data-Efficient Image Transformer (DeiT)-based transfer learning method in speech/image recognition and classification utilizing models pre-trained on image datasets. However, the use of DeiT models, especially those pre-trained on image datasets, has not yet been explored for Valvular Heart Disease (VHD) detection. This paper proposes a transfer learning methodology using the DeiT model pre-trained on image datasets for VHD classification. Additionally, we introduce a hybrid Convolution-DeiT (Conv-DeiT) architecture to further improve classification performance. The Conv-DeiT framework integrates a convolutional block with a Squeeze-and-Excitation (SE) attention mechanism to enhance the channel and spatial information within the input features before processing by the DeiT model. The proposed models were assessed using the Heart Sound Murmur (HSM) database, accessible on GitHub. Experimental results show that the DeiT-based transfer learning approach achieved an overall accuracy of 97.44%. Moreover, our Conv-DeiT method outperformed the DeiT-based transfer learning with an impressive overall accuracy of 99.44%. This study indicates the effectiveness of transfer learning using DeiT models pre-trained on image datasets for heart sound classification. Specifically, our hybrid Conv-DeiT method, which combines the convolutional block and the SE-attention mechanism, demonstrates significant advantages in this context.

Counting fetal movements is essential for assessing fetal health, but manually recording these movements can be challenging and inconvenient for pregnant women. This study presents a wearable device designed to detect fetal movements across various settings, both within and outside medical facilities. The device integrates accelerometer and gyroscope sensors with Internet of Things (IoT) technology to accurately differentiate between fetal and non-fetal movements. Data were collected from 35 pregnant women at Suranaree University of Technology (SUT) Hospital. This study evaluated ten signal extraction methods, six machine learning algorithms, and four feature selection techniques to enhance classification performance. The device utilized Particle Swarm Optimization (PSO) for feature selection and Extreme Gradient Boosting (XGB) with PSO hyper-tuning. It achieved a sensitivity of 90.00%, precision of 87.46%, and an F1-score of 88.56%, reflecting commendable results. The IoT-enabled technology facilitated continuous monitoring with an average latency of 423.6 ms. It ensured complete data integrity and successful transmission, with the capability to operate continuously for up to 48 h on a single charge. The findings substantiate the efficacy of the proposed approach in detecting fetal movements, thereby demonstrating a practical and valuable technology for fetal movement detection applications.

Recent studies have reported the success of linear prediction analysis (LPA)-related features, which are extracted as a short-term spectral feature for replay attack detection due to the advantage of the imperfection in the LPA-based signal produced by recording and playback devices. However, exploiting LPA-based signals is focused on only magnitude-based features and ignores phase-based features. In this paper, we propose two novel LPA-based relative phase features, namely, linear prediction residual-based relative phase (LPR-RP) and linear prediction analysis estimated speech-based relative phase (LPAES-RP). The key idea of both LPR-RP and LPAES-RP is to extract the phase information based on LPA-based signals. In the LPR-RP feature, we modify the relative phase (RP) feature extraction using a linear prediction residual (LPR) derived from the difference between the original/raw speech and LPA estimated speech signal (LPAES) instead of the original/raw speech signal. LPES-RP feature exploits the LPAES signal to replace the original/raw speech signal. Because the trace of the recording and playback device artifacts is the primary evidence for detecting the replayed signal, the advantages of the imperfection of LPR and LPAES are expected to provide efficient phase information for the replay attack detection task. In addition, using the individual LPR-RP/LPAES-RP feature, our proposed features are combined with two standard features, mel-frequency cepstral coefficients (MFCC), constant Q transform cepstral coefficients (CQCC) and the original RP feature, at score level to further improve the detection decision. The performance of the proposed LPR-RP/LPAES-RP feature and combination are evaluated using the ASVspoof 2017 version 2 database. On the evaluation subset, our proposed LPR-RP and LPAES-RP feature achieves a promising improvement over baseline features (MFCC/CQCC). Moreover, the combined systems of LPR-RP, RP, and CQCC obtains an equal error rate of 9.26%.

Recent studies have reported the success of multi-scale convolution neural network (MSConvNet) model for many classification applications due to its powerful ability of exploring multi-scale convolution block to extract multi-scale representations to make a detection. However, a new design based on MSConvNet for pathological voice detection has not been explored. In this paper, we propose RS-MSConvNet, a novel end-to-end MSConvNet model using raw speech for pathological voice detection. The main contribution of the proposed RS-MSConvNet method is to exploit the multi-scale convolution block, followed by spatial-temporal feature block, and fully connected layer as classification. In addition, to further improve accuracy performance, we propose a novel hybrid detection model by integrating the feature extraction ability of the RS-MSConvNet model and the classifier of support vector machine (SVM) method, called RS-MSConvNet-SVM model. The effectiveness of our proposed models is investigated using the TORGO database. The experimental results reveal that the RS-MSConvNet model outperforms other baseline methods in the speaker-independent task. Moreover and as compared to the RS-MSConvNet-SVM model, a further improved accuracy is obtained using the RS-MSConvNet-SVM model. These outcomes exhibit that our proposed models are useful for pathological voice detection.