Lif: A Wizard of Oz Prototype
Testing the viability of a Voice User Interface (VUI) for the Lif IoT device
Lif supports a fulfilling, organic relationship between humans and their house plant by alerting the human when the plant needs care. After analog, 3D model, fabric, movement, and VUI prototyping, we are curious to explore the different interaction modalities that Lif may encompass. In this particular investigation, we will uncover how the VUI component of Lif may be used. Utilizing the Wizard of Oz prototyping method, we tested the Lif VUI on an unknowing participant while our team operated the device's technological component. With this method, we can understand if the Lif VUI is viable or desirable. Whether the technology is feasible is unknown as we are not familiar with crafting a VUI for an IoT device.
Roles
The Wizard of Oz technique is difficult to do alone; for this reason, we had a team help make the test happen. Sara was the facilitator of the test, meaning she interacted with the participant to conduct the test. She also developed the VUI script and created the video to showcase the usability test with the participant. Nadir was the wizard during the test, meaning he operated the VUI while the participant was interacting with Lif. He set up all the voice responses beforehand so that the dialogue would be timely with the live user prompts. Yifan was the scribe and took notes throughout the test. She set up questions to ask before and after the participant completed the VUI tasks and created and recorded the Zoom call.
Prototype
The User
The user for Lif will be plant owners. The most typical use case for it would be to attach to house plants within the home.
Initial Design
To generate the VUI script, Sara used her experience as a plant parent to come up with the most common tasks for plant care. These include:
Watering the plant
Checking for lack of sunlight
Assessing the overall health of the plant
She generated scenario scripts for 5 tasks based on these task requirements, including turning the device on and off.
She then created an overall task flow for the VUI, including common error states and responses.
After refining the scripts through table reads with 3 people, we had a set of dialogues that we could use to build a prototype for Lif.
Wizard of Oz
Due to the constraints of the current pandemic, we had to test our prototype remotely.
To simulate a realistic use case, we used the Wizard of Oz method to mock Lif's functionality and tested it with a participant in the facilitator's home. We used an online text-to-speech tool to build a soundboard of the most common phrases for Lif. These could then be played by Nadir, the Wizard, by sharing his audio in a Zoom session for Sara, the facilitator, and the test participant. Recording this session, with consent, made it easy for Yifan, the note taker, to record her observations.
The test
We designed a simple usability test for Lif to be carried out in a Zoom session. It started by interviewing the participant about her experience with VUI. This questioning line took place to get the participant to have their VUI experience fresh in their memory to compare with our prototype.
Next, we ran through the tasks individually, giving the participant the appropriate scenario for each task. At the end of the test, we unpacked her experience with a set of post-test questions.
1. How certain were you that you completed the tasks?
2. How comfortable did you feel using the voice user interface?
3. Did you know what the VUI was doing when you were interacting with it?
4. Did the terminology of the VUI make sense to you? Why or why not? Do you have any suggestions for changing the VUI’s language?
5. Were you ever confused while using the VUI?
6. Do you think this is a successful way of interacting with the Lif device?
Analysis
The participant completed all assigned tasks of the Life usability test and responded to both pre-task and post-task questions. During pre-task questions, we learned that the participant uses VUI every day but does not consider it useful. She has a Google Home and mainly uses it for music. Instead of asking the participant to think-out-loud, we wanted to immerse the participant in our VUI experience by conducting the post-task questions. All the conversations are not as prolix.
Following are the results we got from post-task questions:
1. She feels certain about the tasks with confirmations from the VUI
2. She feels strange to talk to plant and its responses; She is not used to talking to the Google Home she has
3. Yes, she knows what the VUI was doing during the interactions; It has a very calm voice; it’s clear that it’s coming from the plant
4. Most of the terminology of the VUI makes sense; “Vital” for plants is interesting. She never thinks that it’s connected to plants but regards it related to human vitals; “Sunlight” and “Water Level” make sense
5. If not prompting for every task with specific language, she wouldn't know what to ask the plant. Discoverability needed to be improved.
6. I think so. It’s pretty simple. There is something that I would not remember, like “next time to feed my plant.” I would love to put it on my calendar.
Some unexpected instances occurred during our testing. When the VUI was trying to answer the Sun Level question, it suddenly speeded up due to an unstable internet connection through Zoom. It surprised the participants as she opened up her eyes and raised her eyebrow.
We learned about techniques and error handling related to remote prototyping and remote testing during this challenging time. The VUI seems viable as our unknown participant can navigate the conversation at first use by voice-interacting with the conversational agents. Based on observations of smiles on faces, confirmation of understanding by nodding her head, and positive feedback like “Yes, very calm voice” “It’s clear that it’s coming from the plant” “‘Sunlight and ‘Water Level’ make sense,” we would conclude that the VUI is overall desirable for the participant. Some uses of terminology like “vital” are worth further testing with a larger and more diverse group of participants in the future.
Next steps
- More pre-tasks and preliminary user research
In general, we think that more user research could be done before approaching the user. There are some pre-tasks questions that we could ask to get some quantitative and qualitative results, like “How often do you interact with the VUI in a week?” “In what environment setting do you often get to use a VUI?”
2. More functionality of the VUI.
Additional functionality could be added to the VUI, such as telling the user how much fertilizer the plant needs or dealing with pests or insects. The participant also gave feedback on adding a Calendar function as she suggested that there was something that she would not remember, like “next time to feed my plant.” The VUI could prompt the users and ask if they want to put a feed reminder on their calendar.
3. More discoverable.
According to the participant’s feedback by saying, “If not prompting for every task with specific language, I wouldn't know what to ask the plant,” the discoverability of the functions also needs improvements.
4. More realistic.
From our observations, the participant tried to make eye-contact with the facilitator every time after talking to the Lif. It seems that there is a need for enhanced confirmation for users in addition to the voice confirmation with the VUI. To make the Wizard of Oz more believable and realistic, we would add light to indicate that Lif is talking as a visual confirmation of the received participant’s request.
5. More testing with participants!
Due to time constraints and the class's fast pace, we were only able to involve one participant in our testing. Moving forward, we would love to conduct more VUI testing with participants that may have different user experiences with VUIs. As this test is achieved with a mixed-use of Zoom (for Wizard of Oz sound play and note-taking) and in-person (interactions with the physical device), it would be interesting to try out a completely Zoom remote testing and learn about operating testing virtually with participants.
Authored by Sara Gustafson, Nadir Tareen, and Yifan Lin
