With the advent of Siri and Google Now, conversational voice interfaces are being pushed into the mainstream as a natural way to interact with machines. A huge advantage of using speech technology to interact with computers is that you don’t need a keyboard, mouse, or even a screen to allow you to control things by voice. This means that voice interfaces are especially useful when you’re doing other things with your hands and eyes, such as driving.
It seems reasonable to assume that voice interfaces in cars would be safer for drivers than using manual controls. They save you fiddling about with buttons and looking at displays while driving, so you can keep your hands on the wheel and your eyes on the road. Yet, recent studies have shown that controlling your car by voice can be distracting, and we all know that too much distraction can lead to accidents.
Here are a few reasons why voice interfaces in cars might be more distracting than you’d think:
- Visual display – many voice interfaces do still have a visual display that prints information about what the system is doing, which the driver may need to check.
- Delayed feedback – conversation is relatively slow compared to, say, pressing a button where you get instantaneous feedback. Drivers have to concentrate for longer in order to hear what the system says in response, and to correct it if necessary.
- Lack of context awareness – voice controlled in-car systems have no awareness of where you are on the road, and so don’t necessarily know to be quiet when you’re approaching a busy junction and may need to concentrate. Also, drivers may respond to a conversational interface differently if they are busy concentrating on the road .
- Poor speech recognition – cars are a really noisy environment for doing speech recognition in. Although a lot of work has gone into improving performance in noisy conditions, if the system performs poorly then it will be far more of a distraction than if it works smoothly.
Safer voice interfaces in cars will be ones which can perform well in a noisy car, and adapt quickly to drivers and to road conditions.
 M. Gasic et al, “The effect of cognitive load on a statistical dialogue system“, SigDial 2012.