Speaking to Smarter Cars
If you or I were eavesdropping on someone’s conversation we’d probably be able to tell if the speaker was a man or a woman, if they were young or old, maybe even where they lived or their first language. Computers, on the other hand, have no such ability; in their attempts to understand human speech they often overlook contextual details and user characteristics that people naturally absorb.
Speech recognition is one of the most challenging areas of computer science. While improvements have been slow to materialize, the technology is finally beginning to become a useful tool – as anyone who uses a virtual assistant can tell you. That said, the next major challenge in computerized speech recognition is understanding the contextual details of speech and the personal characteristics of those speaking.
What would this advancement mean for cars? We believe it would be a solid building block for improving the in-car user experience. Let’s look at a couple of cases where a car can help its users once it can “listen between the lines.”
- Juan asks his car to “Enciende la radio”. Since the car doesn’t understand that command, it checks it against a few different languages in its library and determines that the request is in Spanish. It then turns on the radio as requested and automatically switches the infotainment system, speech recognition, and instrument cluster settings to Spanish.
- Olivia uses voice recognition to authorize herself as a new driver of her parent’s sedan. The car recognizes her and auto-sets the “safe-driving” profile her parents have configured for her. Because the car recognizes her voice as that of a young woman, it fine-tunes the speech recognition models using a higher pitch for better accuracy while switching the default satellite radio presets from classic rock to electronic dance music.
- Josephine flies from Montreal to Atlanta and starts up her rental car. The car detects her French accent and prompts her to see if she would like to switch to French. As Josephine is very comfortable in English, she replies “No”. The car then asks if she would prefer metric measurements and she readily answers “Yes”, so the car switches its units to metric.
- As a Brooklyn native, Tony has never before owned a car. His brand-new car detects his New York accent and asks if he would like to enable the “urban native” features – prompted street parking warnings, parking lot pricing display, and automatic traffic avoidance suggestions. Tony knows the city but not how to navigate it by car, so he enables them right away.
These are just a few examples of how the determination of voice characteristics can help improve the user experience in a car. We at Mitsubishi Electric are looking into this along with how voice context can be merged with information from social networks to let the car guide preferences in music, shopping, restaurants, or even the in-car environment. With some appropriate smarts, the car can be an amazing accompaniment to a predictably perfect experience.
Jacek holds a Master’s in Human-Computer Interaction from the University of Michigan, and has a deep background in speech/audio processing technology, as well as voice user interface design. He is responsible for strategic planning activities and coordinating UX projects.