Toward Emotion Recognition From Physiological Signals in the Wild: Approaching the Methodological Issues in Real-Life Data Collection

Physiological data for affective computing in HRI with anthropomorphic service robots: the AFFECT-HRI data set

In human-human and human-robot interaction, the counterpart influences the human's affective state. Contrary to humans, robots inherently cannot respond empathically, meaning non-beneficial affective reactions cannot be mitigated. Thus, to create a responsible and empathetic human-robot interaction (HRI), involving anthropomorphic service robots, the effect of robot behavior on human affect in HRI must be understood. To contribute to this understanding, we provide the new comprehensive data set AFFECT-HRI, including, for the first time, physiological data labeled with human affect (i.e., emotions and mood) gathered from a conducted HRI study. Within the study, 146 participants interacted with an anthropomorphic service robot in a realistic and complex retail scenario. The participants' questionnaire ratings regarding affect, demographics, and socio-technical ratings are provided in the data set. Five different conditions (i.e., neutral, transparency, liability, moral, and immoral) were considered during the study, eliciting different affective reactions and allowing interdisciplinary investigations (e.g., computer science, law, and psychology). Each condition includes three scenes: a consultation regarding products, a request for sensitive personal information, and a handover.

A real-world dataset of group emotion experiences based on physiological data

Affective computing has experienced substantial advancements in recognizing emotions through image and facial expression analysis. However, the incorporation of physiological data remains constrained. Emotion recognition with physiological data shows promising results in controlled experiments but lacks generalization to real-world settings. To address this, we present G-REx, a dataset for real-world affective computing. We collected physiological data (photoplethysmography and electrodermal activity) using a wrist-worn device during long-duration movie sessions. Emotion annotations were retrospectively performed on segments with elevated physiological responses. The dataset includes over 31 movie sessions, totaling 380 h+ of data from 190+ subjects. The data were collected in a group setting, which can give further context to emotion recognition systems. Our setup aims to be easily replicable in any real-life scenario, facilitating the collection of large datasets for novel affective computing systems.

Assessment of the human response to acute mental stress-An overview and a multimodal study

Numerous vital signs are reported in association with stress response assessment, but their application varies widely. This work provides an overview over methods for stress induction and strain assessment, and presents a multimodal experimental study to identify the most important vital signs for effective assessment of the response to acute mental stress. We induced acute mental stress in 65 healthy participants with the Mannheim Multicomponent Stress Test and acquired self-assessment measures (Likert scale, Self-Assessment Manikin), salivary α-amylase and cortisol concentrations as well as 60 vital signs from biosignals, such as heart rate variability parameters, QT variability parameters, skin conductance level, and breath rate. By means of statistical testing and a self-optimizing logistic regression, we identified the most important biosignal vital signs. Fifteen biosignal vital signs related to ventricular repolarization variability, blood pressure, skin conductance, and respiration showed significant results. The logistic regression converged with QT variability index, left ventricular work index, earlobe pulse arrival time, skin conductance level, rise time and number of skin conductance responses, breath rate, and breath rate variability (F1 = 0.82). Self-assessment measures indicated successful stress induction. α-amylase and cortisol showed effect sizes of -0.78 and 0.55, respectively. In summary, the hypothalamic-pituitary-adrenocortical axis and sympathetic nervous system were successfully activated. Our findings facilitate a coherent and integrative understanding of the assessment of the stress response and help to align applications and future research concerning acute mental stress.

Identification of heart rate change during the teaching process

Internet of Things (IoT) technology can be used in many areas of everyday life. The objective of this paper is to obtain physiological functions in a non-invasive manner using commonly available IoT devices. The aim of the research is to point out the possibility of using physiological functions as an identifier of changes in students' level of arousal during the teaching process. The motivation of the work is to find a correlation between the change in heart rate, the student's level of arousal and the student's partial and final learning results. The research was focused on the collection of physiological data, namely heart rate and the evaluation of these data in the context of identification of arousal during individual teaching activities of the teaching process. The experiment was carried out during the COVID-19 pandemic via distance learning. During the teaching process, individual activities were recorded in time and HR was assigned to them. The benefit of the research is the proposed methodology of the system, which can identify changes in students' arousal in order to increase the efficiency of the teaching process. Based on the results of the designed system, they could also alert teachers who should be able to modify their teaching style in specific situations so that it is suitable for students and provides a basis for better teaching and understanding of educational materials. The presented methodology will be able to guarantee an increase in the success of the teaching process itself in terms of students' understanding of the teaching materials.

Emodiversity evaluation of remote workers through health monitoring based on intra-day emotion sampling

Introduction

In recent years, the widespread shift from on-site to remote work has led to a decline in employees' mental health. Consequently, this transition to remote work poses several challenges for both employees and employers. To address these challenges, there is an urgent need for techniques to detect declining mental health in employees' daily lives. Emotion-based health assessment, which examines emotional diversity (emodiversity) experienced in daily life, is a possible solution. However, the feasibility of emodiversity remains unclear, especially from the perspectives of its applicability to remote workers and countries other than Europe and the United States. This study investigated the association between subjective mental health decline and emotional factors, such as emodiversity, as well as physical conditions, in remote workers in Japan.

Method

To explore this association, we conducted a consecutive 14-day prospective observational experiment on 18 Japanese remote workers. This experiment comprised pre-and post-questionnaire surveys, physiological sensing, daytime emotion self-reports, and subjective health reports at end-of-day. In daytime emotion self-reports, we introduced smartphone-based experience sampling (also known as ecological momentary assessment), which is suitable for collecting context-dependent self-reports precisely in a recall bias-less manner. For 17 eligible participants (mean ± SD, 39.1 ± 9.1 years), we evaluated whether and how the psycho-physical characteristics, including emodiversity, changed on subjective mental health-declined experimental days after analyzing descriptive statistics.

Results

Approximately half of the experimental days (46.3 ± 18.9%) were conducted under remote work conditions. Our analysis showed that physical and emotional indices significantly decreased on mental health-declined days. Especially on high anxiety and depressive days, we found that emodiversity indicators significantly decreased (global emodiversity on anxiety conditions, 0.409 ± 0.173 vs. 0.366 ± 0.143, p = 0.041), and positive emotional experiences were significantly suppressed (61.5 ± 7.7 vs. 55.5 ± 6.4, p < 0.001).

Discussion

Our results indicated that the concept of emodiversity can be applicable even to Japanese remote workers, whose cultural background differs from that of individuals in Europe and the United States. Emodiversity showed significant associations with emotion dysregulation-related mental health deterioration, suggesting the potential of emodiversity as useful indicators in managing such mental health deterioration among remote workers.

The Mandarin Chinese auditory emotions stimulus database: A validated set of Chinese pseudo-sentences

Emotional prosody is fully embedded in language and can be influenced by the linguistic properties of a specific language. Considering the limitations of existing Chinese auditory stimulus database studies, we developed and validated an emotional auditory stimuli database composed of Chinese pseudo-sentences, recorded by six professional actors in Mandarin Chinese. Emotional expressions included happiness, sadness, anger, fear, disgust, pleasant surprise, and neutrality. All emotional categories were vocalized into two types of sentence patterns, declarative and interrogative. In addition, all emotional pseudo-sentences, except for neutral, were vocalized at two levels of emotional intensity: normal and strong. Each recording was validated with 40 native Chinese listeners in terms of the recognition accuracy of the intended emotion portrayal; finally, 4361 pseudo-sentence stimuli were included in the database. Validation of the database using a forced-choice recognition paradigm revealed high rates of emotional recognition accuracy. The detailed acoustic attributes of vocalization were provided and connected to the emotion recognition rates. This corpus could be a valuable resource for researchers and clinicians to explore the behavioral and neural mechanisms underlying emotion processing of the general population and emotional disturbances in neurological, psychiatric, and developmental disorders. The Mandarin Chinese auditory emotion stimulus database is available at the Open Science Framework ( https://osf.io/sfbm6/?view_only=e22a521e2a7d44c6b3343e11b88f39e3 ).

Bosses without a heart: socio-demographic and cross-cultural determinants of attitude toward Emotional AI in the workplace

Biometric technologies are becoming more pervasive in the workplace, augmenting managerial processes such as hiring, monitoring and terminating employees. Until recently, these devices consisted mainly of GPS tools that track location, software that scrutinizes browser activity and keyboard strokes, and heat/motion sensors that monitor workstation presence. Today, however, a new generation of biometric devices has emerged that can sense, read, monitor and evaluate the affective state of a worker. More popularly known by its commercial moniker, Emotional AI, the technology stems from advancements in affective computing. But whereas previous generations of biometric monitoring targeted the exterior physical body of the worker, concurrent with the writings of Foucault and Hardt, we argue that emotion-recognition tools signal a far more invasive disciplinary gaze that exposes and makes vulnerable the inner regions of the worker-self. Our paper explores attitudes towards empathic surveillance by analyzing a survey of 1015 responses of future job-seekers from 48 countries with Bayesian statistics. Our findings reveal affect tools, left unregulated in the workplace, may lead to heightened stress and anxiety among disadvantaged ethnicities, gender and income class. We also discuss a stark cross-cultural discrepancy whereby East Asians, compared to Western subjects, are more likely to profess a trusting attitude toward EAI-enabled automated management. While this emerging technology is driven by neoliberal incentives to optimize the worksite and increase productivity, ultimately, empathic surveillance may create more problems in terms of algorithmic bias, opaque decisionism, and the erosion of employment relations. Thus, this paper nuances and extends emerging literature on emotion-sensing technologies in the workplace, particularly through its highly original cross-cultural study.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00146-021-01290-1.

Fear Detection in Multimodal Affective Computing: Physiological Signals versus Catecholamine Concentration

Affective computing through physiological signals monitoring is currently a hot topic in the scientific literature, but also in the industry. Many wearable devices are being developed for health or wellness tracking during daily life or sports activity. Likewise, other applications are being proposed for the early detection of risk situations involving sexual or violent aggressions, with the identification of panic or fear emotions. The use of other sources of information, such as video or audio signals will make multimodal affective computing a more powerful tool for emotion classification, improving the detection capability. There are other biological elements that have not been explored yet and that could provide additional information to better disentangle negative emotions, such as fear or panic. Catecholamines are hormones produced by the adrenal glands, two small glands located above the kidneys. These hormones are released in the body in response to physical or emotional stress. The main catecholamines, namely adrenaline, noradrenaline and dopamine have been analysed, as well as four physiological variables: skin temperature, electrodermal activity, blood volume pulse (to calculate heart rate activity. i.e., beats per minute) and respiration rate. This work presents a comparison of the results provided by the analysis of physiological signals in reference to catecholamine, from an experimental task with 21 female volunteers receiving audiovisual stimuli through an immersive environment in virtual reality. Artificial intelligence algorithms for fear classification with physiological variables and plasma catecholamine concentration levels have been proposed and tested. The best results have been obtained with the features extracted from the physiological variables. Adding catecholamine’s maximum variation during the five minutes after the video clip visualization, as well as adding the five measurements (1-min interval) of these levels, are not providing better performance in the classifiers.

Exploring Unsupervised Machine Learning Classification Methods for Physiological Stress Detection

Over the past decade, there has been a significant development in wearable health technologies for diagnosis and monitoring, including application to stress monitoring. Most of the wearable stress monitoring systems are built on a supervised learning classification algorithm. These systems rely on the collection of sensor and reference data during the development phase. One of the most challenging tasks in physiological or pathological stress monitoring is the labeling of the physiological signals collected during an experiment. Commonly, different types of self-reporting questionnaires are used to label the perceived stress instances. These questionnaires only capture stress levels at a specific point in time. Moreover, self-reporting is subjective and prone to inaccuracies. This paper explores the potential feasibility of unsupervised learning clustering classifiers such as Affinity Propagation, Balanced Iterative Reducing and Clustering using Hierarchies (BIRCH), K-mean, Mini-Batch K-mean, Mean Shift, Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and Ordering Points To Identify the Clustering Structure (OPTICS) for implementation in stress monitoring wearable devices. Traditional supervised machine learning (linear, ensembles, trees, and neighboring models) classifiers require hand-crafted features and labels while on the other hand, the unsupervised classifier does not require any labels of perceived stress levels and performs classification based on clustering algorithms. The classification results of unsupervised machine learning classifiers are found comparable to supervised machine learning classifiers on two publicly available datasets. The analysis and results of this comparative study demonstrate the potential of unsupervised learning for the development of non-invasive, continuous, and robust detection and monitoring of physiological and pathological stress.

Measuring and Modeling the Effect of Audio on Human Focus in Everyday Environments Using Brain-Computer Interface Technology

The goal of this study was to investigate the effect of audio listened to through headphones on subjectively reported human focus levels, and to identify through objective measures the properties that contribute most to increasing and decreasing focus in people within their regular, everyday environment. Participants (N = 62, 18–65 years) performed various tasks on a tablet computer while listening to either no audio (silence), popular audio playlists designed to increase focus (pre-recorded music arranged in a particular sequence of songs), or engineered soundscapes that were personalized to individual listeners (digital audio composed in real-time based on input parameters such as heart rate, time of day, location, etc.). Audio stimuli were delivered to participants through headphones while their brain signals were simultaneously recorded by a portable electroencephalography headband. Participants completed four 1-h long sessions at home during which different audio played continuously in the background. Using brain-computer interface technology for brain decoding and based on an individual’s self-report of their focus, we obtained individual focus levels over time and used this data to analyze the effects of various properties of the sounds contained in the audio content. We found that while participants were working, personalized soundscapes increased their focus significantly above silence (p = 0.008), while music playlists did not have a significant effect. For the young adult demographic (18–36 years), all audio tested was significantly better than silence at producing focus (p = 0.001–0.009). Personalized soundscapes increased focus the most relative to silence, but playlists of pre-recorded songs also increased focus significantly during specific time intervals. Ultimately we found it is possible to accurately predict human focus levels a priori based on physical properties of audio content. We then applied this finding to compare between music genres and revealed that classical music, engineered soundscapes, and natural sounds were the best genres for increasing focus, while pop and hip-hop were the worst. These insights can enable human and artificial intelligence composers to produce increases or decreases in listener focus with high temporal (millisecond) precision. Future research will include real-time adaptation of audio for other functional objectives beyond affecting focus, such as affecting listener enjoyment, drowsiness, stress and memory.

FLIRT: A feature generation toolkit for wearable data

BACKGROUND AND OBJECTIVE

Researchers use wearable sensing data and machine learning (ML) models to predict various health and behavioral outcomes. However, sensor data from commercial wearables are prone to noise, missing, or artifacts. Even with the recent interest in deploying commercial wearables for long-term studies, there does not exist a standardized way to process the raw sensor data and researchers often use highly specific functions to preprocess, clean, normalize, and compute features. This leads to a lack of uniformity and reproducibility across different studies, making it difficult to compare results. To overcome these issues, we present FLIRT: A Feature Generation Toolkit for Wearable Data; it is an open-source Python package that focuses on processing physiological data specifically from commercial wearables with all its challenges from data cleaning to feature extraction.

METHODS

FLIRT leverages a variety of state-of-the-art algorithms (e.g., particle filters, ML-based artifact detection) to ensure a robust preprocessing of physiological data from wearables. In a subsequent step, FLIRT utilizes a sliding-window approach and calculates a feature vector of more than 100 dimensions - a basis for a wide variety of ML algorithms.

RESULTS

We evaluated FLIRT on the publicly available WESAD dataset, which focuses on stress detection with an Empatica E4 wearable. Preprocessing the data with FLIRT ensures that unintended noise and artifacts are appropriately filtered. In the classification task, FLIRT outperforms the preprocessing baseline of the original WESAD paper.

CONCLUSION

FLIRT provides functionalities beyond existing packages that can address unmet needs in physiological data processing and feature generation: (a) integrated handling of common wearable file formats (e.g., Empatica E4 archives), (b) robust preprocessing, and (c) standardized feature generation that ensures reproducibility of results. Nevertheless, while FLIRT comes with a default configuration to accommodate most situations, it offers a highly configurable interface for all of its implemented algorithms to account for specific needs.

An Intra-Subject Approach Based on the Application of HMM to Predict Concentration in Educational Contexts from Nonintrusive Physiological Signals in Real-World Situations

Previous research has proven the strong influence of emotions on student engagement and motivation. Therefore, emotion recognition is becoming very relevant in educational scenarios, but there is no standard method for predicting students’ affects. However, physiological signals have been widely used in educational contexts. Some physiological signals have shown a high accuracy in detecting emotions because they reflect spontaneous affect-related information, which is fresh and does not require additional control or interpretation. Most proposed works use measuring equipment for which applicability in real-world scenarios is limited because of its high cost and intrusiveness. To tackle this problem, in this work, we analyse the feasibility of developing low-cost and nonintrusive devices to obtain a high detection accuracy from easy-to-capture signals. By using both inter-subject and intra-subject models, we present an experimental study that aims to explore the potential application of Hidden Markov Models (HMM) to predict the concentration state from 4 commonly used physiological signals, namely heart rate, breath rate, skin conductance and skin temperature. We also study the effect of combining these four signals and analyse their potential use in an educational context in terms of intrusiveness, cost and accuracy. The results show that a high accuracy can be achieved with three of the signals when using HMM-based intra-subject models. However, inter-subject models, which are meant to obtain subject-independent approaches for affect detection, fail at the same task.