The Demo Site

TALKING to your car and giving it instructions might seem like a wild idea from a sci-fi film – but voice recognition technology is already here and slowly creeping into vehicles.

And there's no need to be daunted by it, as most of us will already use it. Far from being inspired by movies, automated speech commands have been around for a long time. Whenever you call your bank, mobile provider or train company to be given a list of options, it's probably the same technology that's now used in cars. It all started in the domestic sector, and was developed thanks to computer dictation software.

Such programs provided the groundwork for the latest systems, and what worked for them is now being incorporated into vehicles – both with integrated or portable devices. For motorists, these voice recognition and text to speech capabilities will enable them to concentrate on driving rather than on using secondary controls, so they're claimed to be more convenient and safer.

To find out if that's really the case, we consulted experts and tried the latest developments from market leader Nuance. They revealed the hardware is simple – a small but powerful PC, speakers and microphones – yet the development costs are huge. This is particularly the case with built-in systems that have to work straight out of the box – and be competitive when the car reaches the marketplace, years after the concept was devised. Basic controls for stereo or climate functions are prompted by a question requiring a defined answer – often a simple yes or no. The device recognises only such programmed responses, and ignores others. However, with more complicated instructions – such as those for selecting a radio station – speech patterns and the individual sounds of the commands are analysed before being matched with possible options on a database.

Navigation instructions are an important consideration, and that's where voice recognition becomes more difficult. With around 20,000 street names in English alone, the parameters are much wider. Contrary to popular belief, though, the more complicated the address, the easier it is for the device to process. And while accents vary widely, scientists have worked out a method by which the gadget 'learns' how its user pronounces things from the moment it's switched on.
By building a profile, the kit stores the details on its memory, ready for the next time that person instructs the device, normally resulting in a 20 per cent improvement on the default mode. So although at first it might offer unexpected variations on particular words, it becomes increasingly accurate with repeated use.

The next step is what the experts are calling an "open-ended system" – one in which the speech becomes more natural. Rather than the dialogue being limited to what the device can understand, the human will use whatever words they like, resulting in commands such as: "I fancy a Chinese meal – any ideas where I can get one?" At present, such a question would cause confusion. But thanks to constantly learning profiles and improving technology, the kit will 'reply' with a list of nearby restaurants, probably with links to their web pages – and direct you to whichever you choose!

As drivers won't have to take their eyes off the road or their hands off the wheel, the result will be more convenient and safer use of such devices. Being able to search for local data simply by asking the vehicle will eventually lead to voice commands as the standard way to control in-car kit.

It's set for introduction on top-end models being designed right now – which means in only a few years' time it'll be less 'back to the future' and more 'back to reality' for voice recognition vehicles!