What do people do with Alexa in their living room? How do they speak to Siri in their car? Voice interactions are harder for researchers to observe than many other interfaces. They often happen in private spaces, at unpredictable times and might only last a few seconds.
At a recent Research Thing meetup, we got to meet some of the researchers trying to make a dent in this space. Read on to find out what they said, and sign up for our Webcredible voice event if your organisation is thinking of making a foray into voice.
Although there are now a lot of prototyping tools for voice like SaySpring and BotSociety, the BBC team decided to keep things simple and start by having a human reading a script simulate the responses of a voice assistant. They then moved on to using a "sound board" app to play back pre-set answers based on what people had asked for.
Here at Webcredible, we've used both of these approaches too, and it's important to remember you don't need to go high-tech to prototype your idea - voice is something everyone has! Sure enough, the BBC team also found these more than adequate to test their prototype in a real kitchen, and get some great insights:
Voice interactions often happen in private spaces, at unpredictable times and might only last a few seconds
Stuart Reeves and his team from the University of Nottingham have gone one step further, and built custom hardware to monitor not just the direct conversation that people have with an Alexa device, but also what else is being said before and after.
By trawling through hours of detailed transcripts, they were able to establish that voice perhaps still isn't as advanced an interaction pattern as we'd like it to be. Between 30-50% of attempts to interact with an Amazon Echo were a failure, with a voice command either not recognised at all, or returning an incorrect answer.
Fortunately, much of this can be solved by better error messaging - definitely a thing we've always considered at Webcredible when designing not just for voice but for any platform. So when something goes wrong, instead of Alexa saying "Sorry, I don't understand the question" we always aim to steer the user towards relevant commands or questions that produce an outcome.
Finally, Tom Hoy proposed that there are a few different classes of voice-based interactions, each with its own rules. Watching people interacting with voice interfaces in lots of research sessions around the world, he's seen differences between:
So far, it feels like voice is being used a lot for simple commands, and expanding gradually into finding some information or completing some more complex tasks. Tom likened this to having a "5th limb" to support you in daily tasks, or an extension of your mind that has access to much more information. However, we haven't yet reached the point where voice assistants can have long-running conversations around complex subject areas.
This is certainly a useful framework we'll have in mind when we next run workshops with our clients to come up with ideas around voice assistant applications.
While helping our clients build voice apps, we've also run our own research, observing customers' voice interactions in the wild (such as in train stations) and surveying a large sample of voice assistant users in the UK.
Join us in our upcoming Voice event with speakers from Webcredible, Amazon, Virgin Trains and AND Digital to hear what what we found and come up with ideas for using voice to best serve your customers.