Speaker clustering is the task of identifying the unique speakers in a set of audio recordings (each belonging to exactly one speaker) without knowing who and how many speakers are present in the entire data, which is essential for speaker diarization processes. Recently, off-the-shelf deep speaker embedding models have been leveraged to capture speaker characteristics. However, speeches containing emotional expressions pose significant challenges, often affecting the accuracy of speaker embeddings and leading to a decline in speaker clustering performance. To tackle this problem, we propose DTG-VAE, a novel disentanglement method that enhances clustering within a Variational Autoencoder (VAE) framework. This study reveals a direct link between emotional states and the effectiveness of deep speaker embeddings. As demonstrated in our experiments, DTG-VAE extracts more robust speaker embeddings and significantly enhances speaker clustering performance. Our code is available: https://github.com/Toby28/DDSESC.

This paper expands the cascaded network branch of the autoencoder-based multi-task learning (MTL) framework for dynamic facial expression recognition, namely Multi-Task Cascaded Autoencoder for Dynamic Facial Expression Recognition (MTCAE-DFER). MTCAE-DFER builds a plug-and-play cascaded decoder module, which is based on the Vision Transformer (ViT) architecture and employs the decoder concept of Transformer to reconstruct the multihead attention module. The decoder output from the previous task serves as the query (Q), representing local dynamic features, while the Video Masked Autoencoder (VideoMAE) shared encoder output acts as both the key (K) and value (V), representing global dynamic features. This setup facilitates the interaction between global and local dynamic features across related tasks. Additionally, this proposal aims to alleviate overfitting of complex large model. We utilize autoencoder-based multi-task cascaded learning approach to explore the impact of dynamic face detection and dynamic face landmark on dynamic facial expression recognition, which enhances the model’s generalization ability. After we conduct extensive ablation experiments and comparison with state-of-the-art (SOTA) methods on various public datasets for dynamic facial expression recognition, the robustness of the MTCAE-DFER model and the effectiveness of global-local dynamic feature interaction among related tasks have been proven.

This paper investigates the impact of randomly initialized unique encoding of classification label position on the visual masked self-supervised pre-trained model when fine-tuning downstream classification tasks. Our findings indicate that different random initializations lead to significant variations in fine-tuned results, even when using the same allocation strategy for classification datasets. The accuracy gap between these results suggests that the visual masked self-supervised pre-trained model has an inherent preference for classification label positions. To investigate this, we compare it with the non-self-supervised visual pre-trained model and hypothesize that the masked self-supervised model exhibits a self-aware bias toward certain label positions. To mitigate the instability caused by random encoding, we propose a classification label position ranking algorithm, Label Ranker. It is based on 1-D dimensionality reduction of feature maps using Linear Discriminant Analysis and position-rank encoding of them by unsupervised feature clustering using the similarity property of Euclidean distance. This algorithm ensures that label position encoding align with the model's inherent preference. Extensive ablation experiments using ImageMAE and VideoMAE models on the CIFAR-100, UCF101, and HMDB51 classification datasets validate our approach. Results demonstrate that our method effectively stabilizes classification label position encoding, improving fine-tuned performance for visual masked self-supervised models.

Recent psychological and neurological studies suggest that human motor preparation and execution are largely affected by the subjective emotional state. Thus, external emotion stimuli can be a potential tool to enhance the detectability of movement intention from pre-movement neural signals. This article investigated whether emotion-evoking music stimulus could improve the performances of a fully predictive Brain-Computer Interface (BCI) system for movement intention detection. For this purpose, electro encephalo graphical (EEG) signals were recorded from twelve healthy subjects under three emotional conditions: happy, sad, and neutral. The emotions were elicited using external music stimuli while they performed a wrist extension action. Additionally, support vector machine-based offline and pseudo online testing schemes were employed to solve a binary classification problem for determining movement intention from pre-movement EEG. EEG power analysis showed that happy music stimulus resulted in an early occurrence of event-related desynchronization in alpha band compared to other emotional states. Happy emotional stimuli also resulted in comparatively better performances in both offline and pseudo online testing paradigms. The results of this article suggest that external happy music stimulus could enhance the early and accurate detectability of human self-paced movement intention and thus could contribute to the predictive capability of state-of-the-art assistive BCIs.

Prediction of movement intention is of paramount importance to enable volitional control of assistive devices. The performances of current Brain-Computer Interfaces (BCI) are yet to reach the desired degree of accuracy necessary for real-life assistive technologies. Therefore, the prediction of gait intention remains a critical topic of research. This paper proposes an algorithm by leveraging two powerful empirical time-frequency analysis tools: Variational Mode Decomposition (VMD) and Wavelet Synchrosqueezed Transform (WSST). A combination of VMD and WSST was implemented to extract high-quality features in the time-frequency plane from Electroencephalography (EEG) data collected from six healthy individuals. The data were collected while they performed self-paced repetitions of gait initiations and terminations without any external audio or visual cue. The extracted features were later used to train a Support Vector Machine (SVM) classifier with a radial basis kernel to predict the intention to start or stop the gait cycle from gait-related EEG data. The combined VMD-WSST approach reached 83.36±1.75% accuracy, 82.83±2.99% sensitivity, and 83.45±3.59% specificity in starting intention detection. While, in the case of stopping intention prediction, the classification accuracy, sensitivity, and specificity were 81.57±1.70%, 81.14±3.06%, and 82.06±3.28%, respectively. The performances obtained by the proposed methodology were better than or competitive with those obtained by numerous state-of-the-art BCI methodologies. The results of this study show promise in predicting intention to start or stop walking from EEG, which could potentially enable assistive devices to be controlled volitionally.

The hip assistance robotic exoskeleton has been demonstrated as an effective device to assist elderly and disabled people with gait disorders. The assistance efficiency of these devices, however, is less optimized because the parameters in the active impedance control are manually designated. This paper presented a novel assistance control scheme to address the sub-optimal issue. This study poses that the assistance efficiency can be maximized by modifying the mechanical impedance to resonate with the muscle driving force, in which the human-exoskeleton coupling system is approximated with a second-order dynamical system. Based on this, the exoskeleton virtual stiffness is adaptively tuned to make the system intrinsic frequency align with the intended swing frequency. The proposed assistance control scheme demonstrated an increased assistance efficiency than the conventional active impedance control in a simulated study. Experiments that were managed on a newly custom-made hip assistance robotic exoskeleton also demonstrated strong evidence of improved gait kinematics with decreased muscle-skeleton efforts.

In this study, a passive arm-support exoskeleton was designed to provide assistive aid for manufacturing workers. The exoskeleton has two operating states which can be altered using an unique ratchet bar mechanism with two blocks fixed on the ratchet bar. When the upper arm is elevated to the highest poiont, the pawl module will touch the lower block to allow the pawl separated, so that the arm can move freely without any resistance. When the upper arm is depressed to the lowest point, the pawl module will touch the upper block to make the pawl re-engaged, so that the upper arm can be locked at any vertical position. For purpose to improve the ergonomical property, the structural parameters of the exoskeleton were determined by particle swarm optimization. The designed exoskeleton was simulated in the Adams model to investigate its actual performance. A preliminary experimental study was conducted to evaluate the effectiveness of the designed exoskeleton on alleviating users' physical loads in holding heavy tools; the muscular activities on the shoulder muscle groups involved in the weights bearing, elicited by the surface electromyography (EMG) over the shoulder, were significantly reduced from three healthy subjects who carried hand-held tools. The simulation and experiment results show that the designed exoskeleton could effectively relieve the shoulder burden by transferring the bearing load to the waist, where the motion of the arm was not obstructed.

To obtain synchronized gait assistance, this paper presents a new delayless adaptive dual-oscillator (ADO) scheme to address the inherent delay issue. In the ADO structure, a new oscillator is coupled with the primitive one but the phase is adaptively feed-forward compensated. It's remarkable that the compensated phase is determined by the proposed extended phase lag observer, in which both the phase lag and phase leading can be properly estimated and eliminated in the steady and non-steady gait. Moreover, a unified exoskeleton control scheme based on ADO is further proposed to improve the gait segmentation, velocity/acceleration estimation, intention estimation, and assistance generation performances, which further enhances the assistance synergy and reduces the safety risks. Experimental results demonstrate better alignment assistance and consequently reduced muscle efforts with ADO-based assistance control.

Accurate acceleration acquisition is a critical issue in the robotic exoskeleton system, but it is difficult to directly obtain the acceleration via the existing sensing systems. The existing algorithm-based acceleration acquisition methods put more attention on finite-time convergence and disturbance suppression but ignore the error constraint and initial state irrelevant techniques. To this end, a novel radical bias function neural network (RBFNN) based fixed-time reconstruction scheme with error constraints is designed to realize high-performance acceleration estimation. In this scheme, a novel exponential-type barrier Lyapunov function is proposed to handle the error constraints. It also provides a unified and concise Lyapunov stability-proof template for constrained and non-constrained systems. Moreover, a fractional power sliding mode control law is designed to realize fixed-time convergence, where the convergence time is irrelevant to initial states or external disturbance, and depends only on the chosen parameters. To further enhance observer robustness, an RBFNN with the adaptive weight matrix is proposed to approximate and attenuate the completely unknown disturbances. Numerical simulation and human subject experimental results validate the unique properties and practical robustness.

Near-Infrared Spectroscopy (NIRS) signals have been widely used to monitor hemodynamic changes in clinical and psychological investigations as well as human system interfacing such as brain computer interfacing (BCI) for gait and rehabilitation. However, the estimation of hemodynamic changes might be blurred due to the presence of motion artifacts in a moving human in the loop system, which should be removed for a more accurate estimation. To register the motion information more accurately, a wearable wireless NIRS cyber sensor system was developed capable of registering motion-related signals from a multisensory integrated Inertia Measurement Unit (IMU) placed close to the NIR optical sensor. Although multi-axis accelerometer, gyroscope and magnetometer signals that are highly correlated to the motion at the optical sensor may provide a good estimation of the motion artifacts in the NIRS signal, the motion fusion algorithms might provide more accurate estimation of motion artefacts in the NIR signal by overcoming the intrinsic limitations of individual sensors such as imprecision and drifts. This study was purposed to determine whether the combination of motion fusion algorithm-based signal and individual sensor readings from IMU could provide a more accurate correction of the motion artifacts in the NIRS signal. The results revealed that the signal-to-noise ratio (SNR) increased significantly when motion fusion signals were used in the estimation and removal of the motion artifacts. The results suggest that the motion fusion algorithm can provide a more accurate estimation and removal of motion artifacts and thus, supporting a better detection of hemodynamic changes.

In neurorehabilitation systems, early detection of gait intention with a lower false-positive rate and higher sensitivity is a critical factor which dictates the successful operation of the exoskeleton or assistive system. Traditional supervised learning algorithms often fail to reach these goals due to the lack of ability to adapt to the diverse interaction between the external environment and the human user. A threshold regulation approach might be a simple yet significant addition to increasing the adaptability of the intention detection system. In this paper, the performance of a pseudo-online BCI system in asynchronous detection of human gait intention from movement-related Electroencephalography signals is investigated. Seven healthy individuals participated in the study who performed self-paced cycles of gait initiation and termination in multiple trials. A custom-made eight-channel EEG system was used to capture the movement-related neural signals while a pair of in-sole pressure sensors and an Electromyography (EMG) sensor were used for time locking the actual moments of movement onset and termination. A wavelet transform based method along with Hjorth parameters, was employed to extract informative features which were then used to train an SVM-RBF classifier with threshold regulation for asynchronous detection of gait intention. A high true positive rate, low false-positive rate, and very low latency were achieved by the proposed methodology. The results demonstrate the feasibility of the proposed framework in building a human-in-the-loop neurorehabilitation system.

In this study, a human-chair model was developed as the basis for a wearable chair design. A prototype chair, HUST-EC, was fabricated and evaluated. Employing the optimization under an inner point penalty function, an optimized simulation of the operating mode with the lowest chair height was implemented. The solid models were established by using the finite element analysis program embedded in Solidworks, which revealed that the support from the designed chair was steady to the user. An electromyography (EMG) test platform has been developed, consisting of four EMG sensors, a MATLAB-based acquisition software, and a loaded vest. Four healthy subjects participated in the evaluation experiment, in which EMGs were collected from the muscle groups of rectus femoris, biceps femoris, vastus medialis, and vastus lateralis under different loads and chair angles. The experimental data demonstrate that (1) the HUST-EC can greatly reduce muscle activation at a variety of loads and bending angles; (2) under the same load, the muscle activation decreases slightly with an increased bending angle; and (3) at the same bending angle, muscle activation increases slightly with an increased load. The results show that the designed chair can effectively reduce the physical burden in workers and may improve work efficiency.

The uncertainty due to performance variations of IC chips and resource sharing on multi-core platforms have significantly degraded the predictability of real-time systems. Traditional deterministic approaches based on the worst-case assumptions become extremely pessimistic and thus unpractical. In this article, we address the problem of scheduling a set of fixed-priority periodic real-time tasks on multi-core platforms in a probabilistic manner. Specifically, we consider task execution time as a probabilistic distribution and study how to schedule these tasks on multi-core platforms with guaranteed Quality of Service (QoS) requirements in terms of deadline-missing probabilities. Moreover, it is a well-known fact that the relationship among task periods, if exploited appropriately, can significantly improve the processor utilization. To this end, we present a novel approach to partition real-time tasks that can take both task execution time distributions and their period relationships into consideration. From our extensive experiment results, our proposed methods can greatly improve the schedulability of real-time tasks when compared with existing approaches.

Cloud data centers are widely employed to offer reliable cloud services. However, low resource utilization and high power consumption have been great challenges for cloud providers. Moreover, the rapid increase in demand for affordable cloud services magnifies the obstacles for proficient resource management policies. In this paper, we investigate how to improve resource utilization and power consumption in cloud data centers when delivering services with statistically guaranteed Quality of Service (QoS). We assume that the service provider hosts different types of services, each of which has request classes with different QoS requirements. Different from the traditional approaches that distribute workloads with different QoS levels on different Virtual Machines (VMs), we introduce an approach to pack requests of the same service type, even with different QoS requirements, into the same VM, and to remove potential failure requests in time to improve resource usage and energy cost. We formally prove that our algorithm can statistically guarantee QoS conditions in terms of deadline miss ratios. We develop a cloud prototype to empirically validate our proposed methods and algorithm. Our experimental results demonstrate that our approach can significantly outperform other traditional approaches in terms of QoS guarantees, power consumption, resource demand and electricity cost.

Multicore platforms are becoming the primary choice to achieve high performance in today's embedded system design. However, under the current IC technology, the dramatic increase in power density has made the thermal issue a critical concern in design of multicore systems. In this paper, we study the problem on how to determine if a periodic dynamic voltage and frequency scaling (DVFS) schedule for a multicore platform is thermally feasible in satisfying a given peak temperature constraint. To solve this problem, we first develop a novel analytic method to quickly calculate the temperature at an arbitrary time instant, which can achieve orders-of-magnitude speedups over the HotSpot simulator. We then present an approach to pinpoint the peak temperature of a given periodic multicore DVFS schedule. Finally, we develop three methods to check the thermal feasibility of an arbitrary schedule. We formally prove the fundamental principles and validity of our proposed methods and use simulation results to demonstrate their effectiveness.

Thousands of accidents and fatalities occur each year due to drowsy and fatigued drivers who choose to operate motor vehicles despite their reduced level of alertness. Actively monitoring Steering Wheel Movements (SWM) has been an important and well documented method for the detection of drowsy driving. Despite the efficacy of the SWM method, it has yet to be widely deployed widely on motor vehicles as a practical means for individual early detection due to the cost prohibitive nature of current methods as well as complexity of installation and implementation. Due to these limitations, potentially lifesaving methods based on SWM monitoring have not been widely implemented. This paper assesses the efficacy of a proposed low-cost accelerometer-based method of SWM monitoring by extracting various SWM parameters and using the extracted data to train machine learning algorithms. Experimental results suggest that the use of adequately trained Support Vector Machines with Accelerometer-based SWM can be a valuable tool in the detection of drowsy driving and the reduction in death and injuries.

In this paper, a novel scheme called blind interference neutralization is provided in the 3- cell interference channel with a shared instantaneous relay. The shared relay not only receives signals from sources, but also sends signals to destinations with a processing matrix. With the proposed scheme, each destination is able to pick up its own desired signal without encountering inter-user interference (IUI). In particular, the sources are blind in the sense that no channel state information (CSI) is required for transmit beamforming. Numerical simulation shows that the proposed scheme, compared with others, can increase the sum rate performance.

A human-robot cooperative approach to reliable planning and execution is presented. The human-robot system consists of three components: human user, wheelchair robot and the noninvasive brain-computer interface (BCI) which can represent limit types of user's intention patterns based on EEG signals, with insufficient decoding accuracy and time delay. To achieve efficient navigation and positioning under condition of decoding uncertainties of the BCI, three cooperative modes are proposed for specific situations based on trade-off of robot's autonomy and user's flexibility. The coding protocol in each mode is elucidated in detail, and strategies of mode switching are developed. To achieve continuous and smooth motion, a look-ahead visual feedback is applied, so that the user can adjust the intention and/or actively correct extraction error of the BCI before the robot reaches current path node, and consequently, reliable planning and execution are ensured. The effectiveness of the strategies is evaluated by simulations.

The power change of electroencephalographs (EEG) rhythms is an indicator to investigate neuronal activation underlying EEG electrodes. We developed a software routine for analysis and visualization of spatiotemporal patterns of EEG oscillations: event-related desynchronization (ERD) and event-related synchronization (ERS). Further, the spatiotemporal representation of task-related coherence (TRC) was also proposed for investigation of interregional neuronal connectivity. As concurrent artifacts and volume conduction effect may result incorrect or spatially blurred estimation of EEG features, we have embedded data preprocessing functions to reduce these effects including: correcting eye movement-related artifact, changing reference methods and spatial filtering undesirable neuronal activity. These functions were subsequently applied to study EEG time-frequency patterns during self-paced movements in normal volunteers.