The Impact of Virtual Human Vocal Personality on Establishing Rapport: A Study on Promoting Mental Wellness Through Extroversion and Vocalics

Abstract

Virtual humans are employed in various contexts, including mental health interventions, to encourage users to adopt healthy behaviors. Establishing rapport with users can enhance the effectiveness of these virtual agents. One way to build rapport is by matching the personality between the virtual human and the participant, which may elicit a similarity-attraction effect, shown to increase trust and likeability in interactions. Despite the potential of computer-generated voices to convey personality through the manipulation of vocalic properties, prior research has primarily focused on non-vocal aspects of personality. To address this gap, we conducted an online study that altered a virtual human’s vocalic properties to represent high or low extroversion, with a focus on the role of rapport in promoting mental wellness. In this study, a virtual human provided information on stress-reducing mental-wellness practices to 165 participants. Our findings suggest that synthesizing vocalic properties to resemble a low extroversion level can enhance the persuasiveness of virtual humans and improve rapport, as indicated by participants’ self-reported intention to engage in mental-wellness practices.

1. INTRODUCTION

Virtual humans have been used by researchers to elicit more honest responses in healthcare interventions [12], provide information to patients [4], promote access to mental wellness information [31], and interview people to help identify psychological symptoms [18], among other purposes. These virtual human interventions are designed to promote healthier behaviors, such as medical screening [41], or improve educational outcomes [36]. Virtual humans must employ a variety of persuasion tactics to convince users to adopt healthier behaviors or perform the requested activities. The Computers Are Social Actors paradigm allows researchers working with virtual humans to explore principles normally reserved for human-to-human interactions and apply them to virtual humans [22]. As such, many of the persuasion tactics used by virtual humans have their beginnings in other research fields.

One such field of research is understanding the impact of personality on similarity-attraction. Similarity-attraction refers to the tendency of individuals to be attracted to others who share similar characteristics, such as attitudes, values, and interests [7]. Research has shown that personality-based similarity-attraction extends to human-computer interactions [21]. Although computer-generated speech has been shown to effectively synthesize different personality types [15, 20], there has been little research on the impact of using it with a virtual human to elicit a similarity-attraction effect.

Human-to-human interactions of people with similar personality traits have been shown to increase trust and likability between participants [19], which are essential components in establishing rapport. Establishing rapport is crucial in various social settings, including mental health interventions[5]. Similarity-attraction can serve as a foundation for building rapport between individuals or between a user and a virtual human. In mental health contexts, a strong rapport can enhance communication, promote user engagement, and improve the effectiveness of interventions. By leveraging similarity-attraction through matching extroversion levels in virtual humans, our study aims to better understand the role of rapport in promoting mental wellness.

In this study, we aim to investigate how establishing rapport between the virtual human and the participant, through matching extroversion levels using vocalics based on the similarity-attraction concept, influences promoting mental wellness. By portraying extroversion in the virtual human’s computer-generated voice, we examine the role of rapport in encouraging positive mental health outcomes and the adoption of stress-reducing coping exercises. This approach allows us to explore the potential benefits of utilizing vocalics to create a sense of connection and understanding between users and virtual agents in mental health interventions.

In doing so, this study seeks to answer the following questions:

• RQ1: Does matching the level of extroversion portrayed by a virtual human’s computer-generated voice to the participant’s level of extroversion impact how similar participants feel to the virtual human?

• RQ2: Does matching the level of extroversion portrayed by a virtual human’s computer-generated voice to a participant’s level of extroversion impact the participant’s intention to use stress-reducing coping exercises?

In order to answer these questions, we created an online, asynchronous, and between-subjects study using two virtual humans. Each virtual human portrayed either a high or low level of extroversion. The virtual humans presented information to participants on stress-reducing mental-wellness practices. The effectiveness of similarity-matching of personality using only vocalics was measured using participants’ self-reported similarity to the virtual human and intention to use coping exercises.

For our study, two hypotheses are examined based on these research questions:

• Hypothesis 1: Participants will rate the virtual human as more similar to themselves when vocalic extroversion levels are matched between the virtual human and the participant, compared to when the extroversion levels are unmatched.

• Hypothesis 2: Participants will report an increased intention to use the proposed stress-reducing coping exercises when vocalic extroversion levels are matched between the virtual human and the participant, compared to when the extroversion levels are unmatched.

This study contributes to the field of virtual humans by connecting the research topics of personality and similarity-attraction, vocalics in computer-generated voices, and mental health applications. Research has shown that personality can elicit a similarity-attraction effect, which can influence participants’ feelings of similarity, likability, and trust in human-to-human and human-computer interactions. Research has shown that vocalics can be employed to synthesize personality using computer-generated voices, offering a potential approach to enhancing communication in mental health contexts. By applying vocalics to express personality traits in virtual humans, we can potentially improve mental health outcomes for users interacting with these virtual agents.

However, the application of vocalics in expressing personality with virtual humans for rapport-building in mental health interventions has been understudied. Our study aims to address this gap by investigating the impact of matching extroversion levels in computer-generated voices with those of the participants, examining its influence on perceived similarity to the virtual human and the participants’ intention to use stress-reducing coping exercises. Our findings demonstrate that altering the computer-generated voice to portray low extroversion levels using vocalics can enhance the effectiveness of virtual human interventions in mental health, contributing to a more comprehensive understanding of rapport-building methods in mental health interventions.

2. RELATED WORKS

Our work builds upon prior research in the areas of personality and similarity-attraction, vocalics in computer-generated voices, and virtual humans in mental health. In this section, we provide a cohesive discussion of these related works, identifying the research gaps and illustrating how our study aims to address them.

2.1. Personality and Similarity-Attraction

The role of personality in effective communication is well-known [25]. The Big Five structure, or OCEAN model, classifies personality into five aspects: openness to experience, conscientiousness, extroversion, agreeableness, and neuroticism [9]. People often find others with similar personality traits more likable. This is called the similarity-attraction paradigm [21, 29].

This concept also applies to human-to-computer interactions. This is known as the Computers Are Social Actors paradigm [20, 21, 29]. Our work explores the similarity-attraction effect in virtual humans with high and low levels of extroversion. We examine how this influences participants’ feelings of similarity, likability, and trust in the virtual human in a mental health context.

2.2. Vocalics and Computer-Generated Voices

Some studies have shown that voice interfaces that match users’ personality traits are more likable and trustworthy [6]. Others have found that matching perceived extroversion levels in computer-generated voices leads to higher attractiveness, credibility, and knowledge ratings [20]. However, these studies did not explore matching personality traits in the context of mental health and promoting mental wellness activities.

One primary difference between virtual humans and non-embodied voices is the level of anthropomorphism. Anthropomorphism affects participants’ social responses and perception of personality in human-computer interactions [11, 26]. This distinction highlights the importance of further investigating the influence of vocalics and personality matching in the context of virtual humans designed for mental health interventions.

2.3. Virtual Humans in Mental Health

Virtual humans are non-judgmental partners [12] that can be accessed at users’ convenience and adapted to various scenarios. They have become popular in mental health assessment and promotion. Studies have shown that rapport-building virtual humans can encourage self-reporting of mental health symptoms among military personnel [18]. They have also shown that factors like gender and spoken accent influence the perceived expertise of virtual humans in mental wellness support [10].

Another study has examined matching participant personality with Cialdini’s persuasion strategies to increase the intention to perform positive mental wellness activities [40]. This study found that using likability and reciprocity persuasion strategies with a virtual human was most successful at improving participants’ intention to perform the mental wellness activity. The study found differences in the success of different persuasion strategies between participants with high and low levels of extroversion and recommends tailoring persuasion strategies to match the participant’s personality.

Despite these advancements, there are still research gaps. For example, the impact of vocalics in virtual humans and their influence on participant engagement in mental wellness activities has not been thoroughly explored. Our research aims to improve the design of virtual humans that promote positive mental health behaviors by investigating how the personality aspects of vocalics affect participant engagement in mental wellness activities.

3. MATERIALS AND METHODS

3.1. System Design

This study used the Virtual Interviewer Platform (VIP), designed for the rapid development of conversational virtual humans by subject domain experts [39]. The user interface of the system is illustrated in Figure 2. This platform enables participants to interact with a virtual human using multiple-choice input options that follow a branching script-based dialog system. The platform provides real-time speech for the virtual human that is synchronized with lip animations. The platform allows clients to access virtual human interactions through most modern web browsers at a time and location of the participants choosing.

Figure 2:

Virtual human interaction and user interface.

3.2. Virtual Human Design

Two virtual humans were designed for this study to promote the use of stress-relieving techniques that can improve mental wellness. The virtual human dialog was enacted using a conversational script. These virtual humans varied in the portrayed level of extroversion (high or low) through computer-generated voice vocalic characteristics.

Speech for the virtual human was generated from the conversational script using the Google Text-to-Speech service and altered using Speech Synthesis Markup Language (SSML) parameters. The vocalic parameters used were based on research by Nass et al. on matching extroversion levels of a disembodied voice created with a computer-generated voice [20]. For the high extroversion (extrovert) virtual human, the speech generated was altered to increase the speech rate to 216 words per minute and a base pitch of 140Hz. The low extroversion (introvert) virtual human had the speech generated altered to decrease speech rate to 184 words-per-minute, a base pitch of 84Hz, and the volume lowered by 15% (−1.41db). The voice used for this project was Google en-US-Standard-C (FEMALE). Participants were not informed of the variation in computer-generated voice for the virtual human with which they interacted.

A positive correlation has been shown between patient-physician race concordance and intention to adhere to health-related advice [35]. As most Amazon Mechanical Turk workers are white [14], a light-skinned virtual human was used for the intervention. Prior research has shown a preference for female virtual humans for mental wellness interventions [10] and computer-based learning [23], so a female virtual human was used for the intervention (see Figure 2). The virtual human did not use animation-based non-verbal behaviors during the intervention.

Both virtual humans used the same light-skinned female agent that was designed using Adobe Fuse with a background environment developed in the Unity Real-Time Development Platform. The same conversation script was used for both virtual humans. The conversation script provided simple stress-relieving techniques for the interaction. Both virtual humans provided audible speech output through computer-generated voice altered with SSML tags as previously described. Participants interacted with the virtual human using radio buttons and open-ended text input. Conversation flow was not altered for open-ended text input and only involved simple branching based on user radio button choice selection (e.g., the virtual human responded, “I’m doing good too.” if the participant selected “Good. How are you?” versus moving to the next conversation level if the participant selected “I’m doing okay.”). The virtual humans were both in a seated pose in an office setting and used an idle breathing animation. The virtual humans identified themselves with the name “Taylor” to participants.

3.3. Conditions

This research was conducted to evaluate the effectiveness of matching the vocal extroversion levels to influence people’s intention to practice positive mental health behaviors. The between-subject variable was the level of extroversion portrayed by the virtual human through vocalic cues, with participants interacting with either an extrovert virtual human or an introvert virtual human. The participants’ level of extroversion, derived from a subset of the Big Five personality traits, was also treated as an independent variable. During analysis, participants were divided into either introvert or extrovert groups based on the mean value for all participants. Participants with extroversion values below the mean were placed into the introvert, or low-level extroversion, grouping. Participants with extroversion values above the mean were placed into the extrovert, or high-level extroversion, grouping.

3.4. Participants

Participants for our study were recruited from Amazon Mechanical Turk (MTurk) in the first quarter of 2022. Participants were all MTurk Master workers located in the United States. Participants were paid $5.10 US for the 20-minute study based on national minimum wage rates and guidelines from the university Institutional Review Board (IRB). A total of 176 participants were recruited. There were 10 participants unable to complete the intervention, and one participant failed the attention check. The data of 165 participants were used in the analysis.

Participants reported age range was from 24 to 68 years, with a mean age of 43.35 years. Participants reported their gender as male (n = 93), female (n = 70), and non-binary (n = 2). Participants’ races were White (n = 144), Asian (n = 9), Black (n = 5), and more than one race (n = 7). Most participants reported their native language as English (n = 164), and one participant reported their native language as not being English (n = 1).

3.5. Participant Personality

Participants were divided into two groups based on responses to the Big Five personality extroversion subset pre-survey to examine the impact that vocalic extroversion level has on persuasion. The mean value for participants’ responses to the extroversion survey was 25.97. Participants below the mean of 25.97 were placed in the Introvert group (n = 87), and participants above the mean were placed in the Extrovert group (n = 78). Table 1 provides an overview of matched-unmatched personality conditions.

Table 1:

Personality Trait Conditions

Condition	Virtual Human Personality	Participant Personality
Matched Introvert (n = 40)	Introvert	Introvert
Unmatched Introvert (n = 47)	Extrovert	Introvert
Matched Extrovert (n = 34)	Extrovert	Extrovert
Unmatched Extrovert (n = 44)	Introvert	Extrovert

Literature suggests that personality levels should be grouped by comparing within the sample population [27], rather than comparing with the general population. However, scores falling between 25 and 35 for the extroversion personality subset are typically regarded as indicative of a moderate level of extroversion [1]. Although the mean score of 25.97 for this MTurk sample population is on the lower side, it still falls within the moderate range for extroversion.

3.6. Study Design

The study was online, asynchronous, and between-subjects. The study consisted of five stages: informed consent, pre-survey, virtual human interaction, post-survey, and debriefing. During the pre-survey, participants’ extroversion levels and views on and intentions to use mental health coping strategies were collected. After the end of the pre-survey, participants were randomly assigned to one of the two virtual human characters, using either extrovert vocalics or introvert vocalics. Participants interacted with the virtual human they were assigned to, which took approximately 5-minutes to complete. During the post-survey, participants’ views on and intentions to use mental health coping strategies and demographic information were collected. Participants were informed of the nature and purpose of the study during the debriefing and asked if they wished their data to be included in the study.

3.7. Context

The study was completed entirely online by MTurk Master workers using their personal computing devices and Internet connection. Given this, some aspects of the study could not be controlled, such as the participants’ surrounding environment, screen size, or display resolution during the interaction with the virtual human.

We did not run a manipulation check on the voices in this study as the study used explicit parameters in generating the low and high level extroversion voices. These vocal parameters were shown to portray the appropriate levels of extroversion in previous research [20].

Following the parameters from previous research [20], the volume for the introvert virtual human was lowered by 15%. However, as the volume of the device used during the virtual human interaction could not be set to a controlled level, this likely had no impact on the perception of the introvert virtual human voice.

The study used no additional persuasion strategies between subjects. The study used no facial animations beyond synchronizing lip and mouth movements with the words spoken by the virtual human, nor did it use hand or body movements beyond an idle breathing animation. Each of these factors that were excluded from the study design has been shown to increase the effectiveness of persuasion strategies with a virtual human [24, 32, 36, 40].

3.8. Measures

We used two sets of measures to evaluate the impact of the virtual human intervention on participants. The first set of measures assessed the level of similarity-attraction between participants using three questions shown in Table 2. Participants answered each question using a 7-point rating scale immediately following the intervention. Questions related to Like and Trust were included as research has shown that these two factors are key in determining the level of similarity-attraction [7].

Table 2:

Questions related to the perception of the virtual human and rating scale range

Reference	Question	Scale
Similar	How similar do you feel that you are to Taylor?	not at all - very much
Like	How much do you like Taylor?	not at all - very much
Trust	How much do you trust Taylor?	not at all - very much

The second set of measures assessed the effectiveness of the intervention using the four questions shown in Table 3. Participants answered each question using a 7-point rating scale both before interacting with the virtual human (pre-survey) and immediately following the intervention (post-survey). Changes in participant pre- and post-survey responses were analyzed as outlined in the Data Analysis section below.

Table 3:

Pre- and post-survey comparison questions and rating scale range

Reference	Question	Scale
Beneficial	My use of coping strategies to improve my mental health is	not beneficial - beneficial
Approval	Most people who are important to me approve of my use of coping strategies to improve my mental health	unlikely - likely
Confidence	I am confident that I can use coping strategies to improve my mental health	false - true
Intent	I intend to use coping strategies to improve my mental health	unlikely - likely

3.9. Data Analysis

An Aligned Rank Transform (ART) procedure was used for the analysis of the pre- and post-survey independent data [38] to understand the interactions between the participant extroversion groups and the portrayed extroversion level for the virtual human. Although the nature of the rating scale’s ordinal structure does not conform to standard parametric testing, the ART procedure builds a model of the data to allow parametric tests to be used on nonparametric data. An ANOVA was performed on the ART transformed data for the 7-point rating scale questions of participants’ perception of the virtual human shown in Table 2 and the pre- and post-survey comparison shown in Table 3.

4. RESULTS

4.1. Participant Personality Similarity-Attraction to the Virtual Human

Mean and standard deviations of participants’ perception of a feeling of similarity to the virtual human based on participant’s personality are presented in Table 4. We observed no significant interaction effect between matching participant and virtual human extroversion level and ratings of similarity.

Table 4:

Mean and standard deviation for participant perception of the virtual human

	Matched Introvert M (SD)	Unmatched Introvert M (SD)	Matched Extrovert M (SD)	Unmatched Extrovert M (SD)
Similar	3.05 (1.97)	2.98 (1.61)	3.91 (1.76)	4.39 (1.88)
Like	4.58 (2.03)	4.64 (1.95)	5.79 (1.25)	5.84 (1.49)
Trust	4.63 (1.78)	4.68 (1.63)	5.56 (1.35)	5.59 (1.54)

We asked participants to rate their level of trust in the virtual human and how much they liked the virtual human, as these metrics have been shown to increase with similarity-attraction [17, 20]. While responses to the trust and like questions were slightly higher than ratings for similarity, there was no significant interaction effect between the participant extroversion level and the virtual human vocalic extroversion level.

4.2. Virtual Human Voice

We examined the mean differences between the pre- and post-survey questions for Beneficial, Approval, Confidence, and Intent, as shown in Table 5. We observed that pre- and post-survey participant ratings for the Beneficial (Introvert Voice M = 0.65, SD = 1.06; Extrovert Voice M = 0.49, SD = 1.24), Confidence (Introvert Voice M = 0.76, SD = 1.02; Extrovert Voice M = 0.48, SD = 1.15), and Intent (Introvert Voice M = 0.60, SD = 1.01; Extrovert Voice M = 0.48, SD = 1.01) increased more when paired with a virtual human using the low extroversion voice. Analysis of the pre-post measures for participants’ rating showed a statistically significant interaction effect for the introvert voice in both participant personality conditions for the Beneficial (F = 17.0, p < 0.01), Confidence (F = 6.43, p = 0.01), and Intent (F = 5.27, p = 0.02) questions. There were no significant statistical interactions for the Approval pre-post measures.

Table 5:

Mean and standard deviation of pre- and post-survey comparison question differences

	Matched Introvert M (SD)	Unmatched Introvert M (SD)	Matched Extrovert M (SD)	Unmatched Extrovert M (SD)
Beneficial	0.90 (1.22)	0.67 (1.52)	0.24 (0.61)	0.43 (0.85)
Approval	0.48 (1.06)	0.72 (1.26)	0.26 (0.62)	0.25 (0.72)
Confidence	1.08 (1.16)	0.57 (1.21)	0.35 (1.07)	0.48 (0.79)
Intent	0.80 (1.07)	0.64 (1.01)	0.26 (0.99)	0.41 (0.92)

4.3. Participant Personality

We observed that the measures of the participant’s perception of the virtual human ratings for Similar (Introvert M = 3.01, SD = 1.78; Extrovert M = 4.18, SD = 1.84), Like (Introvert M = 4.61, SD = 1.98; Extrovert M = 5.82, SD = 1.38), and Trust (Introvert M = 4.66, SD = 1.69; Extrovert M = 5.58, SD = 1.46) were lower for introvert participants. A significant interaction effect exists between participant extroversion level and participants’ rating for Similarity (F = 17.2, p < 0.01), Like (F = 19.0, p < 0.01), and Trust (F = 14.6, p < 0.01).

Lower scores were also observed for introvert participant responses to the pre-survey questions for Beneficial (Introvert M = 5.01, SD = 1.65; Extrovert M = 6.22, SD = 0.93) and Intent (Introvert M = 5.05, SD = 1.55; Extrovert M = 6.12, SD = 1.06). A significant interaction effect exists between participant extroversion level and participants’ rating for Beneficial (F = 32.1, p < 0.01) and Intent (F = 26.1, p < 0.01).

5. DISCUSSION

This study sought to address two research questions: (RQ1) Does matching the level of extroversion portrayed by a virtual human’s computer-generated voice to the participant’s level of extroversion impact how similar participants feel to the virtual human? and (RQ2) Does matching the level of extroversion portrayed by a virtual human’s computer-generated voice to a participant’s level of extroversion impact the participant’s intention to use stress-reducing coping exercises? The study found that matching a virtual human’s vocalic extroversion level with a participant’s extroversion level did not result in a similarity-attraction effect or impact the effectiveness of the mental health intervention. Instead, the participant’s own extroversion level was the primary factor influencing their ratings.

The implications of these results and possible future research directions are discussed below.

5.1. Voice and similarity-attraction

We set out to explore if synthesizing an introverted or extroverted personality through altering the vocalics of a computer-generated voice would elicit a similarity-attraction effect in participants. The self-reported measure of Similarity by participants showed minor differences, with mean values of 3.05 for matched introverts, 2.98 for unmatched introverts, 3.91 for matched extroverts, and 4.39 for unmatched extroverts. We found that the level of extroversion synthesized through a computer-generated voice did not indicate a similarity-attraction effect in this virtual human-led mental health intervention.

One likely cause is due to the fact that human voices vary widely in pitch and speech rate regardless of the level of extroversion of the speaker [16], which may result in a lack of similarity between the computer-generated voice and the participant’s voice. Additionally, Lee et al. found that while pitch and speech rate are often associated with extroversion, these vocalic cues are not always reliable indicators of personality traits [16].

Existing research studies demonstrate that the similarity-attraction effect can contribute to enhancing adherence to treatment [3, 30], improving patient satisfaction [30], and increasing engagement [37] in health-related interventions. In light of these findings, our study aimed to investigate whether manipulating vocalics could generate a similarity-attraction between participants and virtual humans to improve the outcome of mental health interventions. Although our study did not find a significant effect, it underscores the need for future research to explore alternative approaches to leverage vocalics in order to optimize the impact of such interventions.

5.2. Participant personality plays a strong role in the impact of persuasion techniques

The results of this study indicate that participant personality had an interaction effect with the perceived Similarity, Likability, and Trust in the virtual human, with mean self-reported measures of Similarity being 3.01 for introverts and 4.18 for extroverts, Likability at 4.61 for introverts and 5.82 for extroverts, and Trust at 4.66 for introverts and 5.58 for extroverts. Additionally, an interaction effect was observed between participant personality and ratings for intention to perform coping strategies and their perception of the benefits of coping strategies to mental health, where mean pre-survey ratings of Intent were 5.05 for introverts and 6.12 for extroverts and Beneficial at 5.01 for introverts and 6.22 for extroverts. Extroverts also tended to rate the virtual human more positively than introverts.

One possible explanation for some of the higher ratings from extroverts is the more social nature of this personality type [28]. As virtual humans are designed to look and behave similarly to a human, an extrovert’s social nature may make them more inclined to treat the virtual human in a positive manner [13], which could account for higher ratings of liking and trust of the virtual human.

As these results and previous research indicate [8, 33, 40], people’s personalities matter when being influenced to perform positive, healthy behaviors. Researchers and developers designing virtual human mental health interventions should be mindful of the extroversion level of their participants. To improve the effectiveness of a virtual human intervention, we recommend adjusting the messaging used in the interaction to better match a participant’s personality.

5.3. Virtual human voice in mental health interventions

The results of our study did not show any interaction effect for matching participant personality with virtual human personality for measurements of Similar, Like, and Trust reported post-interaction with the virtual human. However, an interaction effect was shown for the introvert virtual human voice producing an increase in the Beneficial (introvert voice M = 0.65; extrovert voice M = 0.49), Confidence (introvert voice M = 0.76; extrovert voice M = 0.48), and Intent (introvert voice M = 0.60; extrovert voice M = 0.48) question post-survey scores.

A possible explanation for the positive effects of the introvert voice is the similarity in vocalic qualities to those used by human mental health professionals. Mental health professionals tend to speak with an empathetic (lower pitch [2]), softer (lower volume), and more confident (lower pitch [34]) voice that is similar on a vocalic level to the introvert voice used in this study: lower pitch, speech rate, and volume.

These results differ from the findings of Lee et al. that suggested the use of an extroverted voice in situations where the participant personality is not known [15]. However, their study used a disembodied voice under the context of a book buying website to elicit social presence among participants. These differences in results suggest that the context of the interaction has an impact on participant treatment of the voice being used, or more precisely, that the voice should match the domain of the message being delivered.

5.4. Summary and Implications

In this study, we investigated the impact of matching extroversion levels in a virtual human’s computer-generated voice with participants’ extroversion on perceived similarity and intention to use stress-reducing coping exercises. The main findings are as follows:

1. No similarity-attraction effect: Matching vocalic extroversion levels did not result in a similarity-attraction effect in virtual human-led mental health interventions.

2. Participant personality is important: Participants’ personalities influenced their perceived similarity, likability, trust in the virtual human, and intention to perform coping strategies.

3. Introverted voice is more effective: The introverted virtual human voice was more effective at persuasion, regardless of participant personality.

4. Context matters: The context of the interaction plays a significant role in participant treatment of the voice being used.

These findings provide valuable insights for optimizing virtual human-led mental health interventions by considering participant personality and the domain of the message being delivered.

6. CONCLUSION

This study investigated the influence of matching extroversion levels in a virtual human’s computer-generated voice with participants’ extroversion on perceived similarity and intention to use stress-reducing coping exercises. Our findings demonstrated no evidence of a similarity-attraction effect in virtual human-led mental health interventions. However, the participant’s own extroversion level played a significant role in their perceived similarity, likability, trust in the virtual human, and intention to perform coping strategies.

Interestingly, the introverted virtual human voice was more effective at persuasion for both extroverted and introverted participants, suggesting that the context of the interaction plays a crucial role in how participants treat the voice being used. Our findings highlight the importance of considering participant personality and the domain of the message being delivered when designing virtual human-led mental health interventions.

Future research should continue exploring alternative approaches to leverage vocalics and other factors to optimize the impact of mental health interventions. Further investigation into the role of context in participant treatment of computer-generated voices could provide valuable insights for tailoring interventions to specific domains and participant profiles. By expanding our understanding of these factors, we can develop more effective virtual human-led mental health interventions that cater to the unique needs and preferences of diverse populations.

Figure 1:

A virtual human used for promoting positive mental health behaviors.

CCS CONCEPTS.

• Human-centered computing → Virtual reality; Interaction design; Human computer interaction (HCI).