For Airbnb users, the act of traveling offers a break from their usual routine. Going on these breaks may be rewarding, but planning this travel involves dealing with a lot of uncertainty, which can reduce motivation and increase stress. These negative feelings become overwhelming right before users' departures—during moments like packing.

An intelligent voice interface aligns the information in a system with the context of its users - helping them gain clarity and get things done. Our team mapped the data from the Airbnb platform into features that would reduce uncertainty and motivate users for their travels. These functions include suggested packing lists, weather notifications, essential reminders, and recommendations.

Unlike regular applications, voice interfaces don't always have the luxury of communicating with visuals. A voice assistant communicates brand primarily through its words and personality - like a character in a story. Our team used the values from Airbnb's existing writing guidelines to develop four principles for Airby's brand voice: helpful, inquisitive, insightful, and direct.

Human conversations don't usually follow a script, but a good voice interface starts with one. Airby's language needed to be efficient and clear to move people through the packing process comfortably. We used this context to develop our conversation script.

This script would be the basis for our final demonstration of Airby's personality and features. We can categorize the interactions with Airby into two types:

• Dialogue: Extended conversations with multiple options and paths (e.g., packing list interactions)
• Queries: Simple questions or commands (e.g., asking for a weather forecast)

Our final demonstration used an overarching dialogue about the packing list with occasional queries about room amenities and weather.

As the lead prototyper, I was responsible for turning our team's script into a functional voice prototype. For this project, I used Dialogflow - a natural language understanding platform by Google. We could use Dialogflow to deploy a prototype of Airby on a Google Home device as a usable voice experience.

While visual interfaces can guide users through a linear process by disabling unnecessary actions, voice interfaces accept all spoken inputs at every stage of a process. When a system can act on any command at any stage, it makes it more likely for a user to get confused and exit the main task.

To solve this, I adjusted the wording in each response to include a leading question at the end. These questions would show users their available options at any stage in a conversation and provide them with the language necessary to trigger the next action.

I deployed an initial build of Airby onto my Google Home Mini and brought it to the classroom for testing. All four people I tested the conversation flow with were able to successfully create and go through a packing list with their voice.

However, all four participants also noted that the spoken dialogue was often slow or monotonous at times - which made the voice interactions less enjoyable. To make Airby's voice sound more natural, I used SSML (Speech Synthesis Markup Language) to adjust the speed of phrases and insert natural pauses between them.

Fantastic job using a persona and naming that explicitly for your character early...it was a charming personality balanced with what is a very functional use case. Great work identifying the opportunities in becoming proactive about packing and addressing travel anxieties. This is the ideal of what voice apps can do.

- Final written feedback on the project from the instructor, Catherine Herdlick

Our team was largely successful in developing a coherent, enjoyable voice experience around packing for Airbnb users. To improve Airby even further, we could explore:

• Creating a chatbot prototype of Airby to test online
• Using APIs to fetch real-world responses for weather and traffic
• Explore other conversation flows (e.g., booking a listing or communicating with a host)