Top

A Practical Guide to Designing Conversational Voice Interfaces

October 8, 2018

Voice delivers the special power of human connection. According to data about how voice assistance is reshaping consumer behavior on Think with Google, 41% of people who own voice interfaces feel as though they are conversing with a friend or another person. These friends assist people in tasks ranging from checking the weather in the morning to switching the lights off at night.

We explored a voice intervention for the task of learning a new language and designed a Voice User Interface (VUI) called Chattie that helps language learners practice real-world conversations and, in turn, improve their vocabulary. When designing Chattie, we used various design methods and iteratively improved the user interface based on our learnings from multiple usability tests.

In this article, we have distilled the different techniques we used and identified six design principles that can serve as guides for designers creating conversational VUIs:

  • Context—Think about context of use.
  • Explore—Hello, is anyone here?
  • Navigation—It’s difficult to find your way in the dark.
  • Error handling—Try to score a home run.
  • Persona—Hoooooman!
  • Iterate—Eat, sleep, test, repeat!

1. Context: Think About Context of Use

Think about the context in which users will use your product. Validate the need for a VUI solution: Why would users need a voice-based interface? Could they use traditional interactions or existing solutions to accomplish the same task? Do they have access to other tools and devices? If so, how might you leverage them to create a unique and improved experience?

For example, a VUI would be good solution for a hands-free context such as cooking. If users have access to a computer or tablet screen, you could combine a VUI with a Graphical User Interface (GUI) that shows them, for example, how to cut a vegetable or toss pasta. On the flip side, a VUI might not be a good solution for entering banking information—a context where privacy is of utmost importance.

2. Explore: Hello, Is Anyone Here?

The domain of Voice User Experience (VUX) is relatively new, and there are still very few players in this space. Therefore, we encourage UX designers to analyze products outside the VUX space when conducting preliminary research, to get a broader perspective on possible solutions.

For example, in addition to examining other VUIs during our competitive analysis, we looked at offline methods of language learning, mobile apps, and even Spotify podcasts for language learning. This improved our understanding of user expectations, what might be helpful features, and gaps in the domain of language learning.

3. Navigation: It’s Difficult to Find Your Way in the Dark

VUIs can be difficult to navigate without a visual user interface for support, so it is important to simplify navigation and reduce the user’s cognitive load. You can do this by revealing only relevant information at each step through a shallow, but clear information architecture. A VUI must present no more than three options at any given point.

For example, only after a user has chosen a topic to talk about does Chattie elaborate by saying, “I will ask you some questions about your <topic>-related experiences. You can ask me for a sample answer or any word meaning. You can also ask me to mark a sentence for you to practice later. Shall we begin?” Presenting this information before the user selected a topic would require the user to remember it after arriving at this point. This would lead to cognitive overload, which can be incredibly frustrating for the user. To reduce friction, the VUI should also give the user the freedom to navigate back to the previous step or to the main menu.

When designing navigation, it is also helpful to consider the purpose of the VUI. For VUIs that have a specific purpose—such as Chattie, which is for practicing conversations—we suggest presenting the major features to the user up front. In contrast, for VUIs that fulfill many purposes, introducing just a few features, then letting the user explore others later works well. For example, people can use Alexa for checking weather, listening to music, and many other purposes.

4. Error Handling: Try to Score a Home Run

Try to cover all potential user responses when creating a conversation dialogue, but also create generic fallbacks for unscripted responses and unanticipated edge cases. It is also important to tailor fallbacks to include an explanation of why the system cannot complete an action to ensure a smooth experience. To add more value to an experience, you might consider offering an alternative that might be helpful to a distressed user.

For example, if a language learner asks Chattie to provide a specific sample answer that is not available, Chattie says, “I’m sorry, I can’t provide this particular sample answer, but I can give you a different sample answer. Would you like that?”

5. Persona: Hoooooman!

The personality and tone of your VUI will guide the user’s response, so it must be well thought out. What visual branding does for an app, nuances in voice such as tone, quality, and pitch do for a voice-based application.

For example, our usability testing showed that language learners wanted Chattie to be more like a friend than a teacher. Based on this feedback, we made Chattie’s tone friendlier, more encouraging, and natural by adding ad-libs such as “Great answer!” and “That dish sounds so yummy!” to our conversation script.

6. Iterate: Eat, Sleep, Test, Repeat!

Keep users in the loop throughout the design process and iterate based on their feedback. Having real-world conversations with people is a quick way of testing sample dialogues, understanding how users respond to prompts, and how they perceive the questions they’re asked.

It also helps to document unanticipated responses from users. For example, during Wizard-of-Oz usability testing, Chattie asked a participant, “What cutlery do you use to eat this dish?” The participant stumbled for a while because she misinterpreted cutlery as calorie. In preparing for the next test, we applied this feedback by identifying words that might be difficult for a language learner to understand. Then, if users stumbled, Chattie offered to explain these words to them without their asking for help.

In Conclusion

With over 40% of adults using voice technology every day, it is imperative that we, as VUI designers, strive to build seamless experiences. These six principles are a step in that direction. Their aim is to guide your process for designing for voice-based interactions. They incorporate learnings from different steps in the process that we followed when designing Chattie, including the following:

  • validating the need of a VUI
  • understanding the current scenario through a literature review
  • competitive analysis
  • surveys and interviews
  • defining the information architecture
  • iteratively scripting the conversation flow
  • validating solutions through Wizard-of-Oz usability testing
  • polishing the persona and tone of the VUI

We hope our principles will help other VUI designers in their explorations of VUX and bring clarity around potential successes and pitfalls. 

Master’s Student at University of Michigan, Human Computer Interaction

Ann Arbor, Michigan, USA

Pallavi BenawriPassionate about solving problems through design, Pallavi transitioned from a Bachelor’s in Biotechnology to a Master’s in HCI. She is fascinated with the personalization and customization of user experiences. Her goal is to create experiences that are not just seamless, but delightful. She has worked as a Product Design Intern at Visa, doing research, design, and usability testing for the Visa Direct team. Previously, she was a UX Design Intern at Testbook, an education technology startup based in India.  Read More

Master’s Student at University of Michigan, Human Computer Interaction

Ann Arbor, Michigan, USA

Puhe LiangPuhe is an HCI master’s student at the University of Michigan and is passionate about multimodal interactions. She is most interested in combining voice interfaces with other interactive user interfaces such as augmented reality (AR), virtual reality (VR), or physical interactions. She has worked at Alibaba DAMO as an HCI Research Intern, designing gesture control for mid-air, 3D interactions. Before studying at the University of Michigan, she worked as a software developer for enterprise IoT products at SAP.  Read More

Master’s Student at University of Michigan, Human Computer Interaction

Ann Arbor, Michigan, USA

Ruchi OokalkarRuchi is currently a master’s student, studying UX Design and Human Computer Interaction at the University of Michigan. Driven to understand underlying problems and design elegant solutions for them, she transitioned from her work as a Front-End Developer at Flipkart.com, an ecommerce system, to becoming a Product Designer at Sumo Logic. She is passionate about crafting seamless experiences for Web, mobile, and voice-based applications. She is part of the leadership for Women in Voice.  Read More

Master’s Student at University of Michigan, Human Computer Interaction

Ann Arbor, Michigan, USA

Sonali TandonSonali is currently a master’s student and is studying Human Computer Interaction at the University of Michigan. By marrying her design and development skills, she is able to create fresh, interactive experiences for the Web, augmented reality (AR), social computing, and voice-based applications. She has worked as a UX Design Intern at MediaMath, doing UX research and design for their next-gen marketing platform. Before moving to the US, she worked as a Senior Software Developer at Akamai Technologies in India, where she developed Web applications for Akamai’s Partner Portal.  Read More

Other Articles on Voice User Interface Design

New on UXmatters