Time to talk

Suffering from hayfever and have a blocked nose? Wind noise at high speeds on the freeway? Ventilation fans turned up to the max? None of that is a problem according to Jens Halfmann, the developer of the voice dialogue system at Audi. 

“The two microphones in the lighting module are top quality and any remaining background noise is minimised via software. The A8 interior has been precisely calibrated for exactly this purpose. And in many cases, it will turn the ventilation down itself for voice control,” says Halfmann.

The new voice control system, which incorporates not only the navigation but also the media, climate control, most of the telephony and some Audi connect services, makes the new Audi A8 a good listener, at the very least. The driver is free to formulate their requests and instructions as they wish – no need for stilted robotic speech in other words. For example, the system understands the sentence: 

“I would like to go to the Allianz Stadium in Sydney.” Converted into data packages, commands are processed in parallel via two routes – onboard and online.

The hybrid solution makes a lot of sense. If, for example, the driver wants to enter a new special destination, the cloud can provide the benefits of its almost unlimited knowledge. If the network connection is good, the answer comes back in less than two seconds. Onboard recognition works even faster, and offers an astonishingly high hit rate, because the new A8 stores a lot of information locally on user preferences. Halfmann’s co-worker, Doreen Engelhardt cites one example: 

“Let’s say I want to call a Jens, and the system knows three people of that name. It will suggest to me the one that I call most often, i.e. Jens Halfmann.” 

Is that machine learning? “No, not yet,” says Halfmann. “The answer is derived intelligently, but still on a rule basis.”

The voice control in the new Audi A8 not only listens very intently, it can also converse with the driver. If necessary, it seeks clarification, enables corrections, offers choices and also accepts being interrupted. If the online connection should be broken while this is going on, onboard recognition takes over. 

During the dialogue, the driver can transfer between different menus. For instance, they can call a contact from the phone book and enter the associated address as a navigation destination.

How safe is customer data in the new Audi A8? 

“Very safe,” says Doreen Engelhardt, “because it is pseudonymised for online recognition, which makes it virtually impossible to trace back to the individual. And we don’t export the phone book to the cloud at all.” 

Does voice control have any gaps? 

“We have nine main languages that can be recognised in the cloud, including Russian and Mandarin. There are also a lot of other languages in the onboard memory,” says Halfmann. “However, what we can’t recognise is strong dialects. If someone says: ‘Awright hen, a wannae gae tae ra fitba. Whit aboot it?’ then it’ll have a problem. Glaswegian is a very specific kind of English.”