Volkswagen knows that there’s a frustrating disconnect over voice commands in the car. 

Through today’s relatively inexpensive speakers and smart devices at home, you’re able to give simple natural-language requests and most of the time it just works. 

But in the car it’s sometimes a different story completely, with vehicles often fumbling tasks like changing the audio source or raising the temperature unless you consult with a secret decoder ring of syntax commands and the order in which they need to be presented. 

Broadly speaking, it doesn’t feel like vehicle voice systems themselves have gotten any smarter or better in nearly a decade. And yet with the charging and preconditioning that becomes part of smart living with an electric vehicle, it’s becoming expected that EVs fit into the voice-driven Internet of Things ecosystem and just work. 

Voice systems are finally poised for a leap, as hinted in the Type 20 Microbus Concept that Volkswagen revealed earlier this month with a rededication of its Silicon Valley facility, called the Innovation and Engineering Center California (IECC).

Volkswagen Type 20 Microbus concept

Volkswagen Type 20 Microbus concept

The first time you encounter the Type 20 Concept up close—a drivable vintage Microbus converted to electric power—you create a VW Car-Net user login, with a hybrid data system storing some essential information with the vehicle but your primary user profile in the cloud. 

After that, the Type 20 can recognize the owner’s voice—even outside the vehicle, where it can respond to natural-language requests to, for instance, recognize the driver and show its charge level on request in a capacity display on the big VW logo.

With its prototype system, fully operational in this one-off concept, the Type 20 can recognize the voices of multiple family members and bring up preferences before the people even get in. 

In urgent need of a manager

Sagnik Dhar, the leader for the software functions team at IECC, where the next-gen voice system was conceived, explained that the key to getting such features to work, with natural language, involves a stage of software-based processing that cars today don’t include: a much more advanced dialog management system. 

Where the data connection for deciphering commands might not be ideal, and multiple accessories and Internet of Things devices are interfacing with each other within the vehicle, dialog managers are the future for tying it all together. The speech recognition would still be used from a secondary vendor like Google or Nuance, Dhar explained, but the dialog management built in-house would make sure that the vehicle understands what is being requested and, perhaps equally relevant, who is requesting it.

“The speech recognition converting the audio into text, that’s something that’s already out there—that’s state of the art—but it’s understanding the intent, that’s something that’s VW’s knowledge,” he said.

It’s also the dialog manager’s job to efficiently know when a voice command would prompt a request to the cloud (asking about a cafe nearby) versus whether local to the vehicle, like the charge level. 

Byton concept, 2018 Consumer Electronics Show

Byton concept, 2018 Consumer Electronics Show

Dialog managers are the missing piece necessary to get features and systems that claim to be able to respond to multiple occupants’ voice commands—as Byton, for instance, has teased in its concepts—to actually work as advertised. Speech processing will recognize the user, but then that gets translated to a user ID, the settings associated with that user, and then specifically what the user is asking for. 

Dhar said that with the dialog manager VW will be able to incorporate “learning traits that cars don’t have yet.” An example would be saying “I’m hungry”—or any natural-language equivalent—and having the system know not only that you’re looking for restaurants but, from your user database, what sorts of restaurants you like. 

Currently, the best option drivers have for that sort of functionality in new vehicles is off-boarding the task to a third-party app via CarPlay or Android Auto—a clumsy proposition in most vehicles if you use more than one app with the system. 

The IECC develops next-generation concepts, and then if they pass muster they’re handed over to partner facilities in Germany. As Dhar points out, it’s for the VW Group as a whole, so it could be developed in Wolfsburg (VW), Ingolstadt (Audi), or Wiessach (Porsche). 

Volkswagen ID Buzz concept, 2017 Los Angeles auto show

Volkswagen ID Buzz concept, 2017 Los Angeles auto show

Dhar hints that vehicle electronics and software is in a cycle of three years from the development of a concept to getting that idea on the road. So maybe, just maybe, we’ll see timelines sync up—and that all-electric reincarnation of the Microbus arrive in 2022 able to understand what we say.