Investigating a two-way-audio query-response command interface with navigation data extraction for driver assistance

Navigation systems for drivers tend to vocalize what they want, when they want. Drivers who need to clarify instructions are often required to read directions in text form, which is a safety risk during real-time control. Instead, we investigate how a two-way-audio command interface might work. Data is easily extracted from text directions so that questions can be answered without knowledge of the GPS location, as long as the system is aware of the most recently vocalized directional cue (and elapsed time since last utterance). Functions envisioned are as simple as scrolling forward or back through the directions, spelling names or pronouncing differently, repeating the road name or directional change, and saying what is next on command. The queries require modest data extraction with potentially great improvement in usability. Queries may further refer to time and the results of some intermittent internet search, but do not attempt to interpret 2d map data. This project is ongoing, and the contribution includes a novel design architecture that puts a second AI assistant in the co-pilot's seat. This design provides domain-specific memory and vocalization assistance on top of the navigational assistance already familiar to the driver: voice AI (vAI) for improved voice UI (vUI). It aims more for a smart TiVo than co-pilot Siri. When navigation assistance and two-way audio dialogue are provided by the same vendor, the audio recognition and current location errors can be significantly reduced, but the current design is intended as a third-party intervention, adding assistance where the original product is lacking and possibly creating pressure for their products to improve.