Detecting and controlling for the confounding effects of speech activity on heart rate variability in daily life measurements

Human physiology is increasingly being monitored in daily life settings across a number of fields including psychology, psychiatry, medicine, and movement sciences. Tools such as wearable technology and ecological momentary assessment enable researchers to study psychobiological phenomena as they unfold in naturalistic settings. Although daily life monitoring boosts ecological validity and can decrease recall bias, it concurrently introduces unique challenges. Unlike in a laboratory setting, physical activity and postural changes cannot be experimentally manipulated in daily life recordings. Such alterations in metabolic demands can create autonomic and neuroendocrine reactivity that overwhelm those created by psychosocial factors of interest (e.g., reactivity due to perceived acute stress). If these confounding factors are not continuously monitored and accounted for, the physiological reactivity that may have resulted due to psychosocial factors cannot be isolated. To date, a confounding factor that has often been disregarded in ambulatory research is speech. Speaking, a recurring everyday activity, requires unique respiratory demands, characterized by shortened inhalation and prolonged exhalation (McFarland, 2001). Speaking has been found to significantly increase respiratory sinus arrhythmia (RSA) amplitude, which is problematic for ambulatory heart rate variability research (see Saygin et al., 2025b for an overview of other autonomic measures impacted by speech activity). What is RSA and why does speech activity constitute an issue for daily life research interested in RSA measurement? Respiratory sinus arrhythmia (RSA) is the heart rate variability occurring in synchrony with the respiratory frequency. Often used to non-invasively index cardiac vagal activity, RSA amplitude is impacted not only by vagal outflow to the heart but also by changes in respiratory patterns (among other factors). RSA significantly increases with lower respiration rate and higher tidal volume, even in the absence of cardiac vagal changes (Quigley et al., 2024). Recent literature showed that a prolonged exhalation relative to inhalation also increases RSA amplitude even when respiration rate and volume are unchanged (Bae et al., 2021; Laborde et al., 2021; Strauss-Blasche et al., 2000; Van Diest et al., 2014), and does not necessarily slow down the heart rate (Strauss-Blasche et al., 2000; Van Diest et al., 2014). Different approaches have been developed in the past to correct RSA for the effects of respiration rate and tidal volume, to get a better index of cardiac vagal control. Inspiration-to-expiration (IE) ratio, however, most noticeably altered during speech activity (i.e., as shortened IE-ratio) has yet to be incorporated into such a correction method. Studies that focus directly on the effects of speech (rather than IE-ratio) on RSA reported similar findings. In response to mentally stressful tasks performed in silence, RSA decreased, signaling the expected cardiac vagal withdrawal. When participants were told to speak whilst continuing the same task, the drop in RSA was either reversed or largely attenuated (Reilly & Moore, 2003; Sloan et al., 1991). A prolonged expiration (as during speech production) may simply be allowing more time for acetylcholine hydrolysis, increasing the RSA amplitude, although the vagal firing to the heart remains unchanged (Eckberg, 2000; Song & Lehrer, 2003). Not controlling for this increase in RSA induced by prolonged exhalation obstructs our understanding of how psychosocial factors really affect the vagal outflow to the heart. If speech is not detected or controlled, a stressful conservation in daily life, for example, can be misinterpreted as an episode with heightened cardiac vagal activity. Aims of the Current Study In the current procedure, participants had their thoracic respiration (via impedance pneumography) and electrocardiography recorded by the VU-AMS Core over two continuous days. For the first testing day only, participants additionally wore a throat microphone (e.g., with an adjustable strap around the throat) and thoracoabdominal respiratory inductance plethysmography wearable (as recorded by Hexoskin Proshirt) until bedtime. The ambulatory monitoring period was preceded by a short laboratory visit during which the participants breathed through a spirometer for later calibration of the impedance and inductance plethysmography. In the lab, participants also completed a procedure for synchronization between the throat microphone and the respiration-recording VU-AMS Core in time. In a previous laboratory study, we validated and compared several speech detection models utilizing wearable-recorded biosignals including upper-sternum placed accelerometry, thoracic impedance pneumography, or (thoracic and thoracoabdominal) respiratory inductance plethysmography (Saygin et al., 2025b). All methods showed great AUC and accuracy, and the upper-sternum placed accelerometer outperformed the other methods. For these open-access speech detection models to be transferred to daily life research, however, they need to be validated in daily life settings. In the current study, we will validate the upper-sternal accelerometer and the impedance speech detection models against an ambulatory throat microphone coupled with the webrtcvad voice activity detection model (developed by Google) over daily life recordings (see Research Question 2). Upon the validation of the speech detection model(s) against the reference device, we will investigate whether the speech and no speech periods indeed differ significantly in their RSA values in the daily life recordings. In a previous ambulatory study, we compared different approaches of correcting for the effects of respiration rate and tidal volume on RSA and identified an optimal approach: tidal volume-normalized peak-valley RSA (RSA/Vt) (Saygin et al., 2025a, under review). For the purposes of the current study, however, we consider RSA/Vt to be a partial respiratory correction as it does not yet account for the effects of IE-ratio on RSA. We will check whether the speech and no speech periods differ in their partially respiratory-controlled RSA (i.e., RSA/Vt). We will then control this RSA metric also for IE-ratio using two plausible approaches (see Analysis Plan), and explore whether the IE-controlled RSA during speech periods are equal to or even less than RSA during no-speech periods (see Research Question 3). We expect the controlled RSA during speech to be equal to or less than that during silence, as speaking increases cognitive demands and subjects one to social-evaluative stress. There is also previous evidence for increased sympathetic reactivity during speech (e.g., increased salivary alpha-amylase and electrodermal activity). In the current study, we rely on an impedance pneumography signal to calculate our peak-valley RSA values and other respiratory variables (e.g., tidal volume, IE-ratio). However, impedance pneumography respiration previously has only been validated in laboratory settings with relatively stationary activities (Ernst et al., 1999; Houtveen et al., 2006). Using the respiratory inductance plethysmography (RIP) data for the first testing day, we will start our analyses by validating impedance pneumography-derived respiratory parameters against those parameters derived using thoracoabdominal RIP (see Research Question 1). In summary, we aim to develop multiple ways to account for the confounding effects of speech on respiratory sinus arrhythmia amplitude in daily life research: by validating privacy-secure speech detection models in the daily life against a gold standard (which could be used in stratified analyses), and by controlling for the effects of inhalation-to-exhalation (IE) ratio on RSA and testing whether this new metric attenuates or even reverses the confounded (elevated) RSA during speech periods. Because impedance has not previously been validated for recording respiration in daily life, we start by validating the impedance pneumography obtained respiration against dual-band respiratory inductance plethysmography. Thus, the three main research questions we pose are: RQ1: “Can impedance pneumography validly estimate respiratory parameters (including respiration rate, tidal volume, inhalation duration, exhalation duration, and IE-ratio) in daily life settings, as compared to thoracoabdominal respiratory inductance plethysmography?” RQ2: “Can we, with high sensitivity and AUC, detect speech presence in daily life settings using the biosignals of upper-sternum placed accelerometry or impedance pneumography?” RQ3: “Is RSA different between speech and no-speech periods in daily life settings, and can this difference be attenuated or reversed when controlling for the effects of tidal volume and IE-ratio on RSA?” References Bae, D., Matthews, J. J. L., Chen, J. J., & Mah, L. (2021). Increased exhalation to inhalation ratio during breathing enhances high-frequency heart rate variability in healthy adults. Psychophysiology, 58(11), e13905. https://doi.org/10.1111/psyp.13905 Bernardi, L., Wdowczyk-Szulc, J., Valenti, C., Castoldi, S., Passino, C., Spadacini, G., & Sleight, P. (2000). Effects of controlled breathing, mental activity and mental stress with or without verbalization on heart rate variability. Journal of the American College of Cardiology, 35(6), 1462–1469. https://doi.org/10.1016/s0735-1097(00)00595-7 Eckberg D. L. (2000). Physiological basis for human autonomic rhythms. Annals of medicine, 32(5), 341–349. https://doi.org/10.3109/07853890008995937 Ernst, J. M., Litvack, D. A., Lozano, D. L., Cacioppo, J. T., & Berntson, G. G. (1999). Impedance pneumography: noise as signal in impedance cardiography. Psychophysiology, 36(3), 333–338. https://doi.org/10.1017/s0048577299981003 Friedmann, E., Thomas, S. A., Kulick-Ciuffo, D., Lynch, J. J., & Suginohara, M. (1982). The