For this study, 28 people were recruited using an on-campus bulletin board system. The sample size was the maximum number of participants that we could recruit at our university campus during a certain period. Participants received a JPY 1,000 prepaid card as a reward for their participation and ranged in age from 18 to 40 years old, with a mean age of 22.79 ± 4.12 years. Of the 28 participants, eight were men and 20 were women. Before the experiment, it was confirmed that none had any cardiovascular disease that might affect their HR measurements, and none had ever hyperventilated or collapsed due to stress.

This study was approved by the Ethics Committee of Okayama University (Approval No. R1909-003; Registration No. UMIN000037759). All participants provided written informed consent. For participants under 20 years old, consent was obtained from a legal guardian.

Participants were divided randomly into two groups—the Breathe app group and the non-Breathe app group—using the permuted block method with a block size of 9 (Table 1). Each group was assigned a series of tasks to complete in 10 minutes before they prepared to give a speech.

This study’s protocol (Figure 1) was based on the Trier Social Stress Test [19]. All participants wore a mobile, single-lead electrocardiograph (duranta; ZAIKEN Co. Ltd., Tokyo, Japan). First, they rested for 5 minutes. To ease their stress, “Air on the G String,” composed by Johann Sebastian Bach, was played as background music. Next, they completed a State-Trait Anxiety Inventory-S (STAI-S) questionnaire. They then performed the assigned tasks for 10 minutes. Next, they were moved to a second room where lighting, a video camera, and a microphone were prepared. Participants stood in front of the camera and were told the following: (1) their speech would be videotaped and rated in real-time by four evaluators in another room based on their speaking style, their use of language, and the contents of the speech; (2) their voice would be recorded, and its frequency would be analyzed; (3) their speech would be rated; and (4) they had to finish their speeches within 5 minutes. They were also given a speech topic. Next, they returned to the first room and were given 10 minutes to prepare the speech. Once they had prepared, they returned to the second room and delivered the speech for 5 minutes. They were asked to continue speaking if they finished delivering their speech early. After the speech was over, they completed the STAI-S questionnaire for a second time. They then returned to the first room and were asked not to talk about the content of the experiment with others. They received a JPY 1,000 prepaid card as a reward for participation. The researcher accompanied them each time they moved to another room.

Participants performed a speech task as their acute stress task. During their 5-minute speeches, participants were instructed to talk about (1) themselves, (2) sales points, and (3) how to become like their ideal role model. The order in which these speech topics were discussed and for what duration were determined by the participants. This speech topic was designed by the authors to be difficult to answer since a simple topic might not induce enough stress for the purposes of the study.

The time series of the RR interval, the root mean square of successive differences (rMSSD), and HR were determined using an electrocardiograph. Its sampling rate was 160 Hz and the electrode types used were bipolar leads. The power of the low-frequency (LF) component and the high-frequency (HF) component were also calculated to obtain the LF/HF ratio.

Text files of the electrocardiogram data were extracted from the electrocardiograph and were recorded in LabChart version 8 (ADInstruments, Dunedin, New Zealand). HRV-related variables such as HR, the LF/HF ratio, and rMSSD were calculated using the LabChart Pro algorithm (HRV module version 2.0.3). The Welch method was used to perform power spectral density analysis. Ectopic beats were removed when calculating HRV.

According to the international guidelines for HRV analysis, at least 2 minutes of recording is needed to accurately measure the LF component, and a sampling rate of 200 Hz or higher is needed to decrease errors [20]. Although the sampling rate of this study (160 Hz) did not satisfy this standard, the recording duration of this study (5 minutes and 10 minutes) did.

Participants completed the STAI-Form JYZ questionnaires [3], which is a version of the STAI questionnaire adapted into Japanese [21], to evaluate the participants’ state anxiety. The lowest possible score was 20 and the highest was 80. In this study, only state anxiety was used to evaluate transient emotional changes after speech delivery.

The Dunnett test was performed to determine the validity of the speech task. The unpaired t-test was used to investigate the validity of the stress-relieving operation and its effects on the degree of nervousness during speech delivery or speech preparation. The paired t-test was used to compare HR of the Breathe app group at rest with that during the stress-relieving operation and STAI-S scores of the non-Breathe app group at rest with those after the speech delivery. All statistical analyses were conducted using GraphPad Prism7 software (GraphPad Software Inc., San Diego, CA, USA). All statistical tests were two-sided; p-values of less than 0.05 were considered to indicate statistical significance.

Whether the speech task induced stress was examined using HR, LF/HF readings, and STAI-S scores. The effectiveness of the Breathe app at reducing stress was assessed by measuring rMSSD and comparing it between the Breathe app and non-Breathe app groups. Since the LF/HF ratio is affected by deep breathing [22], this indicator was determined to be unsuitable for evaluating stress reduction as the participants practiced breathing regulation. Since the rMSSD is largely not affected by deep breathing [23], it was used instead to evaluate stress.

The speech task induced stress. Significant differences in the average HR, LF/HF ratio, and STAI-S were found in the non-Breathe app group between their resting state and speech delivery, and between their resting state and speech preparation (Figure 2).

During speech preparation, the HR and LF/HF ratio of the Breathe app group were lower than those of the non-Breathe app group (Figure 5), which suggests that stress during this phase may have been eased by breathing exercises using the Apple Watch Breathe app. However, no significant differences based on physiological and subjective indicators were found between the groups during speech delivery (Figure 4). We propose the following three explanations for why stress was not alleviated during speech delivery: the stress of delivering a speech was too intense, the order of the tasks mitigated the Breathe app’s effectiveness, and the stress-easing task was insufficient.

The speech task might have induced more stress in participants than cognitive tasks examined in previous studies, such as mathematical tests and the Stroop task. According to a study that analyzed 208 articles related to acute stress, tasks in which participants do not have control over the setting and are evaluated socially tend to induce a high degree of stress [24]. Participants in the present study were informed that their speeches would be recorded and evaluated by other people. Thus, the speech task meets the condition of “being evaluated socially.” However, in earlier studies [15,25–27], participants only performed cognitive tasks and were not socially evaluated. Due to this difference in the experimental setting, the speech task in this study might have induced more stress than cognitive tasks. As such, the breathing regulation task in this study might have alleviated stress for longer if the speech task induced less intense stress.

In the present study, the following procedure was used: the stress-easing task was performed for 10 minutes, then speech preparation was undertaken for 10 minutes, and then the speech was delivered for 5 minutes (Figure 1). The stress-easing effect may not have lasted through the speech preparation step. In future experiments, the order of the procedure should be changed to improve the stress-easing effect. First, participants should prepare a speech for 10 minutes. Then, they should perform breathing regulation exercises for 10 minutes. Finally, they should deliver the speech. Conducting the steps in this order will allow participants to deliver their speeches before the stress-easing effect diminishes. Another solution could be to give participants a speech topic beforehand so they can devise the content of their speeches at home. Thus, they would not need to spend time preparing the speech and could deliver it directly following the stress-easing operation.

The Breathe app task using Apple Watches was conducted for only one session without pre-training. An earlier study found that breathing regulation using haptic interaction without pre-training increased participants’ HRV [27]. However, the stress-easing tasked used in this study might not have been sufficient. Although a significant difference was observed in rMSSD in between the non-Breathe app and Breathe app groups, no significant differences in HR were found between when the Breathe app group performed the breath-regulating task and when they rested (Figure 3), which is contrary to a previous study that found that HR decreased after an HR control session [28]. In other words, stress decreased during the speech task in terms of rMSSD, but no difference was observed in terms of HR. Therefore, although it complicates the procedure, conducting a pre-training session or increasing the number of stress-easing sessions is expected to be effective at reducing stress.

Adding a new group based on earlier studies can also be considered when conducting future research. For example, one study found that playing music to drivers relaxed them [29]. It is possible that having participants listen to music during the breathing task with the Apple Watch, speech preparation, and speech delivery could reduce stress more. However, it might be difficult to play music that speakers respond positively to during speech delivery. Thus, it would be best to use a method for easing stress that only the speaker can perceive, such as breathing regulation through haptic interaction [27].

Making wearable devices worn by participants vibrate lightly before and/or during stress-inducing activities can also be expected to be effective at reducing stress. Earlier studies succeeded in alleviating subjective anxiety during speech preparation and speech delivery using false heartbeat-like vibrations through a wearable device, doppel, or EmotionCheck that were slower than participants’ actual HRs [16,17]. Another study relaxed participants physiologically by sending haptic feedback from their heartbeat [30]. Furthermore, the BoostMeUp app for Apple Watches enabled participants in one study to perform better on mathematical tests by using vibrations that were slower than their actual HRs [15]. These studies indicate that using another wearable device or other applications installed on Apple Watches to send participants vibrations that are slower than their HRs can ease stress.

In conclusion, we propose five solutions. First, since the speech task might have been too stressful for participants, another less stressful task would enable more effective stress reduction for the purposes of the experiment. Second, changing the order of the procedure would also lead to a difference in stress levels. Since the stress-easing effect might have been lost during the speech preparation step, we suggest making speech preparation the first step and then conducting the breath regulation task immediately before speech delivery. Third, conducting a pre-training session or increasing the number of sessions would reduce stress more effectively, as our results imply that the stress-easing procedure was insufficient to reduce stress meaningfully. Fourth, based on previous studies, we suggest playing music or incorporating haptic interaction so that participants can better regulate their breathing. Lastly, we propose sending vibrations to participants’ wearable devices that are slower than their actual HRs based on the findings of multiple previous studies that demonstrated the effectiveness of vibrations.

There are some limitations to this study. First, the number of participants was low and might not have been high enough for an effective analysis. We recruited the maximum number of participants possible during a certain period. Second, some participants might have been accustomed to using an Apple Watch or might have already been comfortable making speeches in public, which we did not consider when recruiting participants. Thus, the results are potentially biased. Third, the experimental setting was different from the setting of actual speeches. In this experiment, subjects made a speech in front of a researcher. However, in an actual speech, the speaker delivers his or her speech before an audience, which consists of many people. Therefore, our results might not be directly applicable to real-life situations. Fourth, all participants recruited for this experiment were young, and the results might not be generalizable to other generations. Finally, the sampling rate of the device used in this study (160 Hz) was lower than the sampling rate recommended in international guidelines for measuring HRV, which may have led to inaccuracies in the HRV analysis.