Talking Back To Your Radio: How We Approached Voice-UI.
I’m fond of saying that voice-UI devices could be the spiritual successor to terrestrial radio. If you’ve ever yelled at your radio, then you know what I mean. Traditionally, the radio has provided not only information and entertainment, but companionship. Some of us experience comfort in knowing that out there, someone else is sharing that experience with us. But imagine that potential companionship taken to the next step. What if you co uld command your radio? What if you could provide feedback that resulted in meaningful changes to your listening?
It’s no surprise to me that voice-UI devices are growing in popularity. In our homes and cars, on our persons, all the big players have an offering. They are all pushing hard to mature the capabilities of their devices.
At NPR, we want to play a role in the emerging VUI shift. We have relationships with many companies in this space, and we want to be certain that public radio is properly represented as these partners move forward.
Finding our voice
Recently we concluded some research on voice-UI for NPR listeners. We wanted an understanding of how our listeners might use NPR on a VUI device. We started with a few questions:
- What features might users most value in a voice-driven context?
- How would the voice device respond in each scenario? (When should the device give a minimal response? When should it provide more verbose feedback? When should it provide no feedback and just perform the action?)
- What should a conversation with NPR over VUI feel like? (When might users expect to provide additional information? What contextual information are users expecting? How much response information is enough?)
For our research, we used a hypothetical voice-only device. We recruited participants who listened to NPR on a mobile device and used modern voice-assistants. We instructed participants to talk to this device in order to gain access to the features they told us they valued. We then performed the device responses and checked them against their assumptions.
We found that participants expected the VUI device responses to match the terseness and tone of the command. People told us that the device should acknowledge all commands, but unless the audio is being changed, the device should not interrupt with a speech response. We found that additional metadata on responses — particularly duration and episode dates — were quite useful.
Acting on the research
We took a very modest initial approach. To start, we wanted to support initial commands that launched a station stream or started NPR One. Here are a few with our hypothetical VUI device, “Albi”:
Our research suggested we take advantage of the strongest features of NPR One: playing programs users like, allowing them to discover new shows, and allowing on-demand access to the newscast. We built a longer list of commands to handle these cases:
Looking ahead
We’re looking at bringing more core NPR One features to VUI platforms. Right now, all these platforms have their own conventions and “mood”. Cortana and Alexa are more casual, Google Assistant is more formal, and Siri is somewhere in the middle. Also, the command structures are different, notably the need to command Alexa to ask NPR One to perform a task.
Ideally, we’d love for NPR One to have a standardized command set and “response voice” across platforms so that our commands and responses sound natural and authentic.
In the meantime, we’d love to hear your thoughts and suggestions. Drop them in the comments or reach out to us on Twitter.
Got Alexa, Cortana, Siri, or Google Assistant? Find us in your app store.
Further Reading
- Here’s NPR’s official release article on our Alexa Skill by my colleague Ha-Hoa Hamano.
- NPR One in the Alexa Skill store