Audi MMI Voice Control

The "Audi MMI Touch Response" infotainment system was released in 2018. Our goal was to create the leading voice control for a car brand in terms of innovation, intelligence and functional range.

When?

06.2015 - 08.2019

What?

VUI design, user research, function management

With?

Team of 6, division of 50+

The Project

The MMI Touch Response is Audi's infotainment system which was officially integrated in the fourth generation of the A8. In many aspects it was a leap in technology - the system featured two screens, touch controls with real haptic feedback, and a speech dialog system which for the first time was capable of processing natural speech instead of only understanding specific keywords.

The functional range of the voice control was also significantly larger than it had previously been: the new technology allowed drivers to enter whole addresses within a single step, listen to online radio stations, regulate the temperature by telling the system that it was too cold, search for restaurants by mentioning that they are hungry, and many more.

The development took place over a timespan of three years.

Goals and Challenges

While the basic functionality was carried over from the previous Audi infotainment system to the MMI Touch Response, from a design point of view, basically everything was designed from scratch: the GUI screens as well as the voice dialog flows. This was especially true for the speech dialog system, which was based on completely new technology and allowed for more natural dialog flows in comparison to earlier versions of the technology. In this new environment, the speech dialog system could be much more than just keyword based controls of the most basic car functions: we wanted to steer it towards becoming a real assistant and thus a serious alternative to touch controls. To achieve this goal, the main issues to tackle were defined as the following:

  • Functional range: The overall functional range on the previous infotainment generation was limited by it's technology, only offering basic, keyword based voice control. The new generation should offer more commands and more areas, giving a feel that almost every function of the car can be accessed by voice.
  • Customer Acceptance: User research had shown that confidence in earlier generations of voice dialog systems was low, due to frequent misinterpretations of their commands by the system. With updated recognizer technology, the goal was to give the user the possibility to utter their commands freely without having to think about the correct wording, thus leading to more trust and acceptance.
Audi MMI
The two touch screens of the Audi MMI Touch Response. (image source: www.audi.de)
Mode of Operation

I was part of the SDS ("speech dialog system") team, a team of 6 people across Spiegel Institut and Audi, responsible for the concepts and design of the voice control. We oversaw the complete development process of the SDS, from the early concepts until it's release, so our main tasks shifted depending on the project status. During that time, I was mainly responsible for the media and radio dialogs, as well as the ones for the climate control and the car system settings. Temporarily, I also helped out with the telephone and addressbook commands.

During the concept phase, we worked in a Scrum environment. Our team was split up across several interdisciplinary Scrum teams, consisting of UX designers, graphic designers, programmers and testers. Each sprint lasted for two weeks. In this phase, my tasks included the follwing:

  • Realization of user studies to develop and evaluate new conceptual ideas early in the process. These were either conducted in-house by ourselves or in cooperation with our market and user research team.
  • Creation of SDS dialog and screen concepts in accordance with the overall operation concept of the MMI touch response system, as well as specification. This task included brainstorming sessions in the team, Design Thinking workshops and solo work, and was an iterative process.
  • Constant exchange with adjacent teams and stakeholders to ensure technical feasibility and consistency of the system. This was achieved by coordinating regular meetings with SDS team members, Scrum team members and representatives of the technical department.

During the development and QA phases, the main tasks shifted towards fine-tuning the concepts and solving issues during implementation as well as testing pre-releases and ticket management up until the final approvement of the software.

The Spiegel SDS team
Team meetings and brainstorming sessions were held regularly. (image source: YouTube)
Solutions

To significantly improve the functional range, we added new areas to the voice control, which were previously unaccessible by voice control: for example, it was now possible to control the A/C including the temperature, seat heating and other functionality. Other, already implemented areas received new commands: the driver could now switch his radio to music genres or stations from different countries, and he could ask to navigate to contacts he had just called. When the system was unable to obey a command (be it because of legal constraints or lacking implementation), we implemented fitting responses, informing the user about other possibilities to perform the desired task.

To further increase customer acceptance, we implemented new ways of voice interaction, like the possibility of overanswering questions and interrupting the system while it was talking. We also incorporated more natural commands like "I am hungry", "I am tired" and "I am cold". Additionally, we improved on the intelligence of the system, integrating algorithms like the proactive proposition of telephone contacts which were called frequently. Overall, we developed the concepts and specifications for more than 250 different commands, expanding the functional scope of the system significantly in comparison to the previous one, with improvements mainly focusing on recognition and system intelligence.

We performed regular usability tests to make sure the concepts were easily understood by the users, and tested new additions to the system via big-scale studies and Wizard-of-Oz tests. To improve the internal workflow and collaboration and learn from our mistakes, we conducted lessons-leaned workshops with the other teams after each project phase.

Reception

The voice control of the Audi MMI Touch Response received many positive reviews. Below are some of the press reviews from the year 2018.

"Mehrere hundert Kommandos werden verstanden, unabhängig von starren Menüstrukturen und gegebenenfalls mit Rückfragen und pfiffigen Algorithmen."

"Leute, isses das? Ja, das isses."

"Beneath the sharp graphics lies a robust system that can match any competitor for sheer functionality [...] Voice controls are likewise intuitive and robust, accepting more natural inputs and responding quickly and accurately to commands."

Zum Anzeige- und Bedienkonzept des Audi A8 gehören [...] auch eine starke Sprachbedienung [...] Dieses System versteht natürlich gesprochene Sprache und ermöglicht so nicht nur die komfortable und komplette Steuerung des Infotainment- und Navigationssystems, sondern auch die Bedienung der Klimanlage (sic!)."