Tuli

 Personas

For our personas, we wanted to explore two different types of low vision users: a more technologically savvy user and less-technological user. Our two personas include vision level, pain points with current voice technology, and what goals in their travel plans.

 Voice Avatar

Our goal for our travel assistant’s avatar was to be unique, welcoming, and pleasing to look at. For our design I went with a rounded shape with a smiling face, and used natural colors and imagery. The tulip flower on the head of the avatar represents the city of Amsterdam, and to create a unique brand for our voice system. In future work, we may expand the assistant to other cities around the world and include a different flower to represent that city.

Intents, Utterances, Slots

In order to design the voice and tone of the voice assistant, we generated a set of intents, utterances, and slots in a spreadsheet. We categorized our dialogues by intent and generated variations on the user utterances for the system to expect, including wake words and exit words. We designed to accomodate the user’s travel needs but also to provide fun facts, language lessons, and recommend unique local attractions. We generated over 2000 utterances for both user and system to design for not only all possible user utterances, but also adding in randomly generated system utterances. Through competitor research on voice assistant, we found that the most interesting and engaging VUI included randomized utterances for particular user flows.

In our error handling, we planned for how the voice system would handle silence, pauses of various lengths, along with natural language reprompting of the user while taking care not to sound too robotic or forceful.

User Flow and Voice Flow

We designed our user flows simultaneously with our intents, utterances, and slots. We noticed that as we started to write out different flows and error handling, we wanted to adjust the utterances of our system to and realized areas where we were lacking intents, causing us to create new flows. This back and forth process led to a final spreadsheet of all our intents, along with a basic user flow involving twelve different steps our user may take during their Tuli experience.

Next, we used the Voiceflow tool, a voice development interface, to implement our user flows. Voiceflow allowed us the ability to create a working prototype of a voice system, including the ability to upload to Amazon Alexa or Google Assistant. We included ear cons to signify conclusions of dialogue, and when Tuli was listening to the user. We also included error handling for when the system did not match the user utterance to a specific flow. In order to increase accessibility, we built in portions of haptic feedback. For example, when Tuli is asked to take a photo, it will have the customization option to vibrate at the point of capturing the photo. To improve our prototype further, we also made use of APIs through Yelp and Google Maps to provide real time ratings and restaurant recommendations, along with directions to those locations via car, bike, and walking.

Screens

For our voice assistant we wanted to include multiple modalities, so the system utilizes both a visual and voice design. We chose colors that were easy on the eyes and performed web browser color contrast testing via the online WCAG test. In the final design, we would allow for different color contrast customization options for the user to choose from.