Volkswagen and SVOX Develop In-Car Voice System with NLU

SVOX, a specialist in embedded speech solutions, is working with Volkswagen Group Research to make natural language understanding (NLU) voice interfaces in the car a reality. The ambitious goal of the “Voice Control” research project is to eliminate the need for specific voice commands to control in-car functions. Instead of being strictly guided through a menu, drivers can use their natural, everyday speech to control their car’s address book functionality.

Although today’s voice control systems do enable drivers to efficiently control in-car infotainment systems without having to take their hands off the steering wheel, SVOX is working with Volkswagen to make in-car voice commands even more natural and intuitive, and therefore easier to use. This “just speak” approach was made possible by the combination of statistical language models based on collections of utterances made by real people, and the unique slot filling dialogue manager from SVOX.

“The aim of our activities is to make it easier to use the speech dialogue systems. Users will not have to learn commands in future,” commented Peter Oel, head of human factors and human-machine interaction at Volkswagen Group Research.  “Free input, such as Uh, I want to call Clara Schumann now, will now easily be recognized.”

The “Voice Control” project leader at Volkswagen Group Research, Gordon Seitz, explains: “Natural language understanding is made possible by a knowledge-based evaluation of the user input.”  The system filters the relevant information from an utterance, regardless of the speakers' wording or word order. A frame-based dialogue system supports the user in effectively achieving the intended action, asking questions where information is missing and providing additional information where necessary.

Users will now be able to freely correct the system by over-answering and even changing their mind during a command: I want to call Clara Schumann at the office – uh no, let’s send her a message instead, and this will still be understood.

“People in the past were put off using speech technologies because of the necessity to learn specific commands,” said Martin Reber, CEO of SVOX. “Through our work with Volkswagen AG, people will now be able to interact with their cars as though they’re speaking with another person, making voice control easier to use and more efficient. The SVOX slot filling dialogue manager has proven its capability to easily develop speech dialogues, contribute to shorter development time and lower cost while at the same time increasing quality of speech dialogues.”