Great voice design required observation of what people actually say when using Knowcast in the field
We were able to design a reliable custom voice assistant for the Knowcast mobile app using a process called “utterance discovery”.
Problem: the PROMISE of voice is you can say anything. The PROBLEM with voice is you can say anything.
Voice control is a 100% open user experience. There are no guideposts, so people can literally say anything. How do we know what people want to say when they want to control Knowcast hands-free?
Solution: listen to what users say when using it. Rinse, repeat
Fun exercise: imagine you’re out on a walk, and you’re listening to a podcast. You get bored of what’s being said so you want to skip ahead a bit. What would you say?
- “Skip 30 seconds”
- “Jump forward”
- “Skip this bit”
- “Go forward”
… or something else?
The answer is: all of the above! In a previous post I covered why I believe behaviour-based design is best — and this is a great example of behaviour-based design in action. We had no idea what people would say, and in fact, nor did they. The phrases they came up with happened on the spur of the moment. This was gold dust.
We learned how people speak naturally with Knowcast, while on the go.
We added a temporary logging feature that allowed us to listen to what people said, what they meant, and what Knowcast interpreted it to be.
We then performed detailed analysis to understand what happened. This routinely involved listening to the recordings ourselves and making a judgment. We then would update Knowcast’s listening settings to better accommodate any changes we wanted to make. This included intelligent defaults. (For example: when someone says “go back” what did they mean? We settled on 15 seconds.)
This contributed to a great voice experience that “just worked”.
But there were more factors that contributed to this. Stay tuned for future posts!