Using a Robot's Voice to Make Human-Robot Interaction More Engaging

Transcription

1 Using a Robot's Voice to Make Human-Robot Interaction More Engaging Hans van de Kamp University of Twente P.O. Box 217, 7500AE Enschede The Netherlands h.vandekamp@student.utwente.nl ABSTRACT Nowadays a robot is becoming more than just a machine, the robot becomes an interaction partner. A human needs to be engaged to interact with the robot. This paper is about an experiment on robot voices in a task-based environment. The goal was to determine the influence of the robot's voice on the way humans are engaged or interested to perform a certain task. This research is contributing to the topic of engagement in human-robot interaction with different voice styles. The participant is asked to perform six small assignments to measure the effects of the different voices; a human-like voice (N=10) and a machine-like or mechanical voice (N=11). There were some significant differences between the two voices, mostly related to the likeability of the robot. The differences between the voices in terms of interest or engagement turned out to be minimal and not significant. Keywords Robot voice, challenging, engagement, human-robot interaction, task interest, robot interest 1. INTRODUCTION Different frameworks for human-robot interaction have been created in the past years [1]. Most of them are trying to improve human-robot interaction by incorporating human behavior and human personality traits in robots [7, 16]. Voice is an important factor in human personality [8], therefore many robots use a human-like voice to interact with humans. Creating a human-like interaction partner has proven to be valuable in human-robot interaction [4, 10] in terms of effectiveness and efficiency [11, 15]. This research will not evaluate human-robot interaction by measuring task effectiveness or task efficiency, but it will evaluate humanrobot interaction by the way humans are engaged [12] while performing a certain assignment. This research aims to provide insight on the relationship between a robot s voice and engagement [12] in terms of interest. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. 18 th Twente Student Conference on IT, January 25 th, 2013, Enschede, The Netherlands. Copyright 2013, University of Twente, Faculty of Electrical Engineering, Mathematics and Computer Science. 1.1 Problem statement Other research [4] focuses on models in which robots mirror the participants behavior to create a more human-like interaction partner. However, most of that research does not focus on engaging a human. This paper focuses only on engaging a participant in human-robot Engaging is related to the concept of interest as it describes attentional and emotional involvement [12]. The goal of this research is to determine the influence of a robot s voice on the level of engagement in terms of interest [12] in human-robot It seems normal to use a more human-like robot to improve human-robot interaction, assuming that the robot is perceived as human-like. However, it is a fact that most robots are far from being like a human. If a human perceives the robot as artificial and not as human-like, it might influence the expectancy of the robot voice. If a robot is perceived as artificial and uses a mechanical voice instead of the human-like voice, does this influence the degree of which humans are engaged to perform a certain task? Are humans more interested in a robot with a mechanical voice instead of a human-like voice? Both questions are important to understand the relationship between humans and robots in human-robot Or in this paper more specific: the relationship between voice and engagement. With the previous questions in mind, the following research questions can be created. 1.2 Research Questions To investigate the relationship between a robot s voice and participant interest towards the robot or task a few questions need to be answered. The main research question is: Does a robot with a humanlike voice make human-robot interaction more engaging than a mechanical voice? Other research questions will be used to find an answer to the main question: - How can we find out if a person is engaged or interested in the robot or task while interacting with a robot? - Does a robot s voice influence the way participants are engaged to perform a task in terms of interest? Which can be divided into two more questions: o o Does a robot s voice influence the way humans are interested in the robot? Does a robot s voice influence the way humans are interested in the task?

2 2. RELATED WORK Research on human-robot interaction is often about creating a lifelike interaction partner. Therefore a lot of research is conducted with a human-like voice only, of which some are mentioned in the introduction. Important related work to this research is conducted by Walters et al. Rich and Sidner also wrote about the concept of engagement in human-robot However, they did not use different voice styles or robot appearances like Walters et al. A paper on robot appearance and personality by Walters et al. [17] investigated people s perceptions of different robot appearances. The research uses the definitions of robot appearance based on the definitions of Gong and Nass [5] and MacDorman and Ishiguro [9]. In this experiment the robot appearances are referred to as human-like or machine-like. This paper describes the relation between interest and robot voices, it does not talk about appearance preferences in general. years with an average of 21,6 years. The second oldest participant was 24 years old. The experiment using the human voice had 10 participants and the experiment using the robot voice had of those participants were students of exact sciences such as Computer Science, Electrical Engineering, Mathematics or a master in the same area. The other student studied Climate & management and was apparently lost.. Because of the large amount of students with a technical background only 4 participants were female. They were equally divided over both conditions. Other research by Walters et al. [18] used different robot voices similar to the experiment described in this paper. There has also been research on the gender of the robot. Siegel et al. [13] conducted research to determine the preferences of males and females on robot gender. The research showed that females generally did not have a preference of robot gender. However, males seem to have a preference of a female robot gender. Research performed by Crowell et al. [3] indicate that the perceived gender of the robot may influence human sexrelated characteristics. 3. METHODOLOGY In this experiment the Magabot robot is used. A photo of the robot is shown in the figure at the right. The robot is small (less than 1 meter tall) and offers a platform on which a laptop can be placed. The robot will be controlled using a Wizard of Oz technique because of communication problems between the robot and Flash. The robot will drive on a platform between two tables, which will be explained in section 3.3. The voice and eyes are programmed in Flash (ActionScript). The robot used a predefined script in the form of a Flash timeline. The timeline was divided into several segments to allow a researcher to control the robot's script. As seen in the figure at the right, the laptop displayed the robot's eyes. The robot eyes and effects such as blinking were used to make the robot more life-like. In the experiment the robot eyes were used to look at objects. Based on the research conducted by Siegel et al. [13], the robot had a female voice. Both the human-like and machinelike voice were female (or female-like) voices. The used voices are discussed in section 3.2. The robot itself had no gender. The gender was determined by the voice of the robot. The robot introduced herself as Jane. 3.1 Participants A total of 21 participants took part in the experiment using a between subjects design. Their age varied between 19 and 31 Figure 1: the Magabot with laptop and eyes used in the experiment. 3.2 Voice Two different female voices were used in the experiment. The robot used a predefined script (in English) to communicate with the participants. Both voices were created with the MARY Text-to-Speech system. The first group of participants (N=10) interacted with a robot using a synthesized human-like female voice. The second group of participants (N=11) interacted with a robot using a synthesized machine-like or mechanical-like female voice. The used voice was in MARY called 'cmu-slt-hsmm'. The second group had a robot filter to create a more machine-like or mechanical voice. 3.3 Experiment Setup The conducted experiment was combined with research about proxemics in human-robot interaction, or more specific the relationship between a robot's voice and proxemics. The experiment was divided in three parts. In the first part the robot tells its name and asks the participant to come closer. This part was needed for the other experiment and is also a good introduction of the robot. The participant walks towards the robot and the robot then tells the participant to do some simple tasks. The participant is asked to take a seat

3 at the table. The robot drives to the other side of the table, making sure the participant is sitting opposite of the robot. For the second part of the experiment, the participant performs six simple assignments. On the table are six cards with letters forming the word thanks and also six (empty) numbered boxes. Three of the cards are faced up and show the letter, the others show the backside which is colored (red, green and blue). The six assignments are small assignments such as move the letter N to box 2, swap the letter N with the letter H or please turn over all colored squares. The start of each of the assignments was triggered by hand using a Wizard of Oz technique. After some assignments the participant tells the robot how many boxes are left empty to enforce interaction with the robot. After moving all cards the robot asks the participant to flip all colored cards and read the word in the boxes. The robot then thanks the participant and drives to the table on the left-hand side of the participant. The third part of the experiment is very small. When the robot arrives at the table it turns around and asks the participant to come closer. This part is also needed for the other experiment. The robot tells the participant there is a questionnaire on the table next to her and asks the participant to fill it in. The experiment has ended. The figure below shows a picture taken from the experiment setup. The table with the six squares is in front of a platform on which the robot drives around. This was necessary because otherwise the robot was too small and the participant would not have been able to see the robot s eyes. The moving assignments are similar to the ones used by Staudte et al. [14]. However this experiment is not based on utterance or gesture, but instead focuses on the two different voices. Figure 2: the experiment setup. 3.4 Questionnaire & video footage A three-page questionnaire was used to determine if a person was engaged or interested in the robot or task. The questionnaire is based on measurement instruments for measuring anthropomorphism, animacy, likeability, perceived intelligence and perceived safety of robots provided by Bartneck et al. [2] to get understanding of how the robot was perceived by the participants. The questionnaire also contains questions on the attention allocation scale provided by Harms and Biocca [6] to find out if the participants were interested in the robot or the task. The questionnaire consists of five parts. The first part is about some general information. The second, third and fifth part contains 32 questions about the experiment on a 5-point Likert scale provided by Bartneck et al. [2]. The fourth part contains 9 questions which are rated on a scale from 1 to 7, of which some are provided by Harms and Biocca [6]. Because the conducted experiment is shared with Rens Hoegen not all questions will be used for providing answers in this paper Most important questions The categories rating of the participant s have some important questions in order to answer the research questions: - artificial / lifelike & mechanical / Organic: questions in the categories anthropomorphism and animacy to determine how the robot's appearance was perceived. - unfriendly / friendly & unpleasant / pleasant: questions in the category likeability to rate the impression of the robot. - incompetent / competent & unintelligent / intelligent: questions in the category perceived intelligence. Another category contains questions to which the participant can disagree or agree. The most important are: - I was interested in the robot. - I was interested in the task. - I remained focused on the robot throughout our - I remained focused on the task throughout our The questionnaire ends with a manipulation check to determine if the difference in voices was noticed. The experiment was also recorded by two different cameras (one full HD camcorder and one fisheye camera) in order to support the questionnaire. Some small screenshots taken from the video footage can be found in section 4.2. The questions of the questionnaire can be found in Appendix A. 4. RESULTS Most of the 41 questions from the questionnaire were used to analyze the relationship between voices and the level of interest. Some questions were not relevant for this research because it was a combined experiment. At first it is important to determine the reliability of the questions, after that the questions will be used to answer the research questions. Finally some screenshots of the video footage is shown to highlight some details. Figure 3: a schematic representation of the experiment.

4 4.1 Questionnaire The categories anthropomorphism, animacy, likeability, perceived intelligence and perceived safety suggested by Bartneck et al. [2] were all used in the questionnaire Reliability The Cronbach Alpha value of each of the categories is listed in the third column of the table below. The second column shows which questions were used in that particularly category. The matching questions can be found in Appendix A. Table 1: Cronbach Alpha values for each category before removing questions. Category Questions (#) Cronbach Alpha Anthropomorphism 1, 5, 9, 13, Animacy 2, 6, 10, 13, 14, Likeability 3, 7, 11, 15, Perc. intelligence 4, 8, 12, 16, Perc. safety 24, 25, Because most Cronbach Alpha values were not very reliable some questions have been deleted. To achieve a Cronbach Alpha value of at least 0.70 in each category the following questions have been deleted: - moving rigidly / moving elegantly & unconscious / conscious in the category anthropomorphism. - apathetic / responsive, stagnant / lively & inert / interactive in the category animacy. - foolish / sensible in the category perceived intelligence. - quiescent / surprised in the category perceived safety. It is important to state that some participants had trouble with the meaning of apathetic and quiescent, all participants were Dutch. This results in the following Cronbach Alpha values: Table 2: Cronbach Alpha values for each category after removing questions. Category Questions (#) Cronbach Alpha Anthropomorphism 1, 5, Animacy 2, 10, Likeability 3, 7, 11, 15, Perc. intelligence 4, 8, 12, Perc. safety 24, Analysis In the tables below the results of the Independent Samples T- tests is shown. In table 3 the results of the test are shown for Bartneck s categories. The results of the questions used in the manipulation check are shown in table 4. The other questions, including the questions taken from Harms and Biocca are evaluated separately in tables 5 and 6. Table 3: Independent Samples T-tests results for Bartneck s categories with both conditions. Category Voice M SD t(19) p Anthropomorphism Animacy Likeability Perceived intelligence Perceived safety Human-like Machine-like Human-like Machine-like Human-like Machine-like Human-like Machine-like Human-like Machine-like Analyzing and comparing the two tables above shows that the categories animacy, perceived intelligence and perceived safety have no significant difference between the human-like voice and the machine-like voice. The category likeability on the contrary does show a significant difference between the human-like voice and the machine-like voice. The category anthropomorphism was approaching significance. Table 4: Independent Samples T-tests results for the manipulation check with both conditions. Question Voice M SD T(19) p Machinelike / Humanlike Unpleasant / Pleasant Disengaging / Engaging Unclear / Clear Human-like Machine-like Human-like Machine-like Human-like Machine-like Human-like Machine-like The questions machinelike / humanlike, unpleasant / pleasant & disengaging / engaging in the manipulation check about the voice only showed minor differences, mostly in favor of the human-like model. The most important significant difference can be found in the question unclear / clear. It showed that the human-like voice was more clear than the machine-like voice and that a difference in voices was noticed. This can also be seen in the video footage, discussed in section 4.2 below. Some 7-point scale questions are excluded in this research, the others are listed below in tables 5 and 6. For each of the questions the Independent Samples T-test is performed and listed below. Table 5: Independent Samples T-tests results for the remaining questions with the human-like voice. Question M SD t(19) p I feel that the robot is interesting to look at

5 I was interested in the robot I was interested in the task I was easily distracted from the robot when other things were going on. (Recoded) I remained focused on the robot throughout our I remained focused on the task throughout our Understanding the robot was difficult. (Recoded *) voices. These pictures will be used in section 6 to support the conclusion. Figure 3: the participant is fixating on the robot after completing an assignment. Table 6: Independent Samples T-tests results for the remaining questions with the machine-like voice. Question M SD t(19) p I feel that the robot is interesting to look at I was interested in the robot I was interested in the task I was easily distracted from the robot when other things were going on. (Recoded) I remained focused on the robot throughout our I remained focused on the task throughout our Understanding the robot was difficult. (Recoded *) Figure 4: her facial expressions indicate that she has difficulties with understanding the robot. The above tables show that the differences between the two voices in terms of interest is not very significant. Thought it is interesting to know that there is a slight difference in focusing on either the task or the robot. * Note that some questions are recoded to make sure that all questions are rated the same way, meaning that 1 = negative, 4 = neutral and 7 = positive. In the question Understanding the robot was difficult. 1 was positive (not difficult) and 7 was negative (difficult), therefore the question is recoded to match its rating with the other questions. 4.2 Video footage In the previous section it became clear that the human-like voice was more likeable and that understanding the robot was slightly more difficult with the machine-like voice. The machine-like voice was rated as more unclear compared to the human-like voice. The video footage showed the exact same results. The pictures below show some information about the Figure 5: the participant is staring at the background, thinking about what the robot just said. 5. DISCUSSION The difference between the two used voices turned out to be not very significant. The most important differences were found in Bartneck's category likeability and the question unclear / clear in the manipulation check. The human-like voice scored higher on the likeability scale and was found more clear than the machine-like voice. The questions in the category anthropomorphism were only approaching significance and showed that the human-like voice was perceived as slightly more anthropomorphic. There was also a minor difference in understanding the robot. The machine-like voice was harder to understand than the human-like voice, which relates to the difference found in the unclear / clear question.

6 The relationship between voice and interest is not significant. In the introduction the question 'If a robot is perceived as artificial and uses a mechanical voice instead of the usual human-like voice, does this influence the way humans are engaged to perform a certain task?' came to mind. The results show that the influence of the voices is minimal. The robot with the human-like voice was perceived only a little more interesting. The results in table 5 and 6 appear to show a contradiction. The human-model scored slightly better on the questions about interest. However, the questions about focus show a minor difference in favor of the machine-like voice. 6. CONCLUSION As mentioned above, some results appear to be contradicting. Though these results are not as significant as hoped, the difference can still be explained. Because the machine-like voice was rated less clear than the human-like voice, some participants had difficulties with understanding the robot. Section 4.2 shows three pictures taken from the video footage. Figure 4 shows a participant with facial expressions indicating it takes some effort to understand the robot. Figure 5 shows a participant thinking about what the robot said. It seems that the participants with the machine-like voice focused more on the robot because of the unclear voice. With the above results, the research questions mentioned in section 1.2 can be answered: This research showed that a robot's voice does not significantly influence the way participants are engaged to perform a certain task. Nor were the participants significantly more interested in the robot or task. The human-like voice in general scored overall better than the machine-like voice. This result is similar to the results found by other researchers mentioned in the introduction. 7. FUTURE WORK In order to completely understand the relationship between voice and engagement, more research needs to be conducted. To improve this research better utilities are necessary. The used Magabot did not have a body (or body movement) and lacked facial expressions. To improve this experiment some aspects of the robot such as embodiment and perceived gender (more than just the voice of the robot) should be implemented. The robot was not quite socially interactive. The video footage could have been of greater importance to this research. Due to time issues, it was only possible to use the video footage as support to the questionnaire. The analysis of the video footage might reveal more interesting facts in a larger study. Some participants asked how the robot knew they were finished with a certain task. Parts of the experiment were conducted using the Wizard of Oz technique, which might have influenced the way the participant perceived the robot. In the future it is necessary to create a robot which responds autonomous based on the actions of the participant. Future work on this research might use a larger group of participants with equally mixed gender and a less technical background. The experiment was also conducted in a taskbased environment, experimenting in a more real-life situation would be more appropriate. 8. ACKNOWLEDGEMENTS I would like to thank Betsy van Dijk and Manja Lohse for providing guidance and solving problems during the research. And I would also like to thank the University of Twente for providing all necessary utilities to conduct the experiments. My thanks also go to Gilberto Sepúlveda Bradford and Rens Hoegen for helping with the robot and the experiments. 9. REFERENCES [1] C. Bartneck and J. Forlizzi. A Design-Centred Framework for Social Human-Robot Interaction. In Proceedings of the Ro-Man2004, Kurashiki pp ACM. DOI: [2] C. Bartneck, D. Kulić, E. Croft and S. Zoghbi. Measurement Instruments for the Anthropomorphism, Animacy, Likeability, Perceived Intelligence, and Perceived Safety of Robots. In International Journal of Social Robotics, Vol 1(1), DOI= [3] C.R. Crowelly, M. Villanoy, M. Scheutz and P. Schermerhorn. Gendered voice and robot entities: Perceptions and reactions of male and female subjects. In Intelligent Robots and Systems, IROS 2009, IEEE/RSJ International Conference, DOI= [4] J. Goetz, S. Kiesler, and A. Powers. Matching robot appearance and behavior to tasks to improve humanrobot cooperation. In Proceedings of ROMAN 2003, the 12th IEEE International Workshop on Robot and Human Interactive Communication, 55-60, ACM. DOI= [5] L. Gong and C. Nass. When a talking-face computer agent is half-human and half-humanoid: human identity and consistency preference. In Journal of Human Communication Research, Vol 33(2), DOI= [6] C. Harms and F. A.Biocca. Internal consistency and reliability of the networked minds social presence measure. In M. Alcaniz & B. Rey (Eds.), Seventh Annual International Workshop ACM. DOI= [7] K. M. Lee, W. Peng, S. Jin and C. Yan. Can robots manifest personality? An empirical test of personality recognition, social responses, and social presence in human-robot In Journal of Communication, 56, DOI= [8] J. A. LePine and L. Van Dyne. Voice and cooperative behavior as contrasting forms of contextual performance: Evidence of differential relationships with Big Five personality characteristics and cognitive ability. In Journal of Applied Psychology, Vol 86(2), DOI= [9] K. MacDorman and H. Ishiguro. The uncanny advantage of using androids in cognitive and social science research. Interaction Studies, Vol 7(3), DOI=

7 [10] B. Mutlu, J. Forlizzi and J. Hodgins. A Storytelling Robot: Modeling and Evaluation of Human-like Gaze Behavior. In Proceedings of HUMANOIDS 06, ACM. DOI= [11] Y. Okuno, T. Kanda, M. Imai, H. Ishiguro, and N. Hagita. Providing route directions: design of robot s utterance, gesture, and timing. In Proceedings of the 4th ACM/IEEE international conference on Human robot interaction, HRI 09, 53-60, ACM. DOI= [12] C. Peters, G. Castellano and S. de Freitas. An exploration of user engagement in HCI. In Proceedings of the affective-aware virtual agents and social robots (AFFINE) workshop, international conference on multimodal interfaces, ICMI 09, ACM. DOI= [13] M. Siegel, C. Breazeal and M. I. Norton. Persuasive Robotics: The influence of robot gender on human behavior. In Intelligent Robots and Systems, IROS (2009) ACM. DOI= [14] M. Staudte and M. Crocker. Visual attention in spoken human-robot In Proceedings of the 4th ACM/IEEE international conference on Human robot interaction, HRI 09, 77 84, ACM. DOI= [15] A. Steinfeld, T. Fong, D. Kaber, M. Lewis, J. Scholtz, A. Schultz and M. Goodrich. Common metrics for humanrobot In Proceedings of the 1st ACM/IEEE International Conference on HumanRobot Interactions, HRI'06, ACM. DOI= [16] M. L. Walters, K. Dautenhahn, R. te Boekhorst, K. L. Koay, C. Kaouri, S. Woods, C. Nehaniv, D. Lee, and I. Werry. The influence of subjects personality traits on personal spatial zones in a human-robot interaction experiment. In Proceedings of COG SCI 2005: Toward Social Mechanisms of Android Science Workshop, 29-37, ACM. DOI= [17] M. L. Walters, D. S. Syrdal, K. Dautenhahn, R. te Boekhorst and K. L. Koay. Avoiding the uncanny valley: robot appearance, personality and consistency of behavior in an attention-seeking home scenario for a robot companion. Autonomous Robots, DOI= [18] M. L. Walters, D. S. Syrdal, K. L. Koay, K. Dautenhahn, R. te Boekhorst. Human approach distances to a mechanical-looking robot with different robot voice styles. In the 17th IEEE International Symposium on Robot and Human Interactive Communication, RO- MAN (2008), DOI=

8 APPENDIX A. RESULTS QUESTIONNAIRE Table A1: first part of the questionnaire containing questions rated on a scale from 1 to 5. Human-like voice Machine-like voice # Questions Mean Std. Deviation Mean Std. Deviation Rate the impression of the robot on a scale from 1 to 5. 1 Fake / Natural Dead / Alive Dislike / Like Incompetent / Competent Machinelike / Humanlike Stagnant / Lively Unfriendly / Friendly Ignorant / Knowledgeable Unconscious / Conscious Mechanical / Organic Unkind / Kind Irresponsible / Responsible Artificial / Lifelike Inert / Interactive Unpleasant / Pleasant Unintelligent / Intelligent Moving rigidly / Moving elegantly Apathetic / Responsive Awful / Nice Foolish / Sensible Quiet / Loud Unhelpful / Helpful Intimidating / Inviting Rate your emotional state on a scale from 1 to Anxious / Relaxed Agitated / Calm Quiescent / Surprised Unsafe / Safe Pressured / At ease

9 Table A2: second part of the questionnaire containing questions rated on a scale from 1 to 7. Human-like voice Machine-like voice # Questions Mean Std. Deviation Mean Std. Deviation Give your opinion on the following statements. (Scale from 1 to 7, strongly disagree to strongly agree.) 29 I feel that the robot is interesting to look at I was interested in the robot I was interested in the task I was easily distracted from the robot when other things were going on. 33 I remained focused on the robot throughout our 34 I remained focused on the task throughout our Understanding the robot was difficult Throughout our interaction I became more familiar with the robot I felt uncomfortable when I was close to the robot Table A3: third part of the questionnaire containing questions rated on a scale from 1 to 5, used as the manipulation check. Human-like voice Machine-like voice # Questions Mean Std. Deviation Mean Std. Deviation Rate your impression of the voice of the robot on the following scales. 38 Machinelike / Humanlike Unpleasant / Pleasant Disengaging / Engaging Unclear / Clear