The effects of different voices for speech-based in-vehicle interfaces: impact of young and old voices on driving performance and attitude

Ing-Marie Jonsson, Nils DahlbäckInterspeech 2009

TL;DR

This study reveals a counterintuitive finding: while drivers, especially older ones, prefer communicating with a young, friendly voice in their car's in-vehicle system, this preference negatively impacts their driving performance. This highlights a critical trade-off between user experience and safety in speech-based automotive interfaces.

What Problem Does This Paper Address?

As in-vehicle information systems increasingly rely on speech for navigation, communication, and control, understanding how interface design affects driver behavior is crucial. While previous research has shown speech interfaces can be less distracting than visual ones, the impact of specific voice characteristics, like age, on driving performance and user attitude has been largely overlooked. This paper specifically investigates whether matching the age of the in-vehicle system's voice to the driver's age influences their attitude and, more importantly, their driving safety. The authors are motivated by the growing prevalence of conversational AI in cars and the potential for poorly chosen voices to compromise safety.

Methods

The researchers employed a driving simulator (STISim) to conduct their experiment. Participants, divided into two age groups (18-25 and 55-75), interacted with a conversational in-vehicle system that had either a young (20-year-old) or an old (76-year-old) female voice. A control group drove without any voice system.

The experiment involved:

  • Pre-driving questionnaire: To gather general information and prior driving experience.
  • Training course: To familiarize participants with the simulator and screen for simulator sickness.
  • Main driving scenario: A varied and realistic 60,000-foot drive lasting approximately 27 minutes, during which the conversational system posed questions at 18 points.
  • Post-driving questionnaires: To assess attitudes (willingness to communicate, affection, trust, dominance, system quality) and perceived influence on driving.

Driving performance was measured by:

  • Accidents: Combined collisions and off-road incidents.
  • Breaking Traffic Regulations: Speeding, running stop signs, and running red lights.
  • Swerving: Lane deviations.
  • Driving Time: Time taken to complete the course.

Datasets

Not discussed in the paper. The study utilized a driving simulator and custom questionnaires for data collection.

Key Insights and Ideas

  • Preference vs. Performance Paradox: Drivers, particularly older ones, express a strong preference for a young, friendly voice, finding it more affectionate, trustworthy, and less dominant than an older voice. However, this preference leads to worse driving outcomes.
  • "The Car is Different!": Traditional similarity attraction principles (liking people similar to oneself) don't fully apply in the automotive context. While drivers might prefer a similar-aged voice in other settings, in the car, a younger voice is favored, possibly due to associations with expertise or efficiency.
  • Conversational Systems as Secondary Tasks: The nature of conversational systems, especially when initiated by the system, can be a significant distraction. The system needs to be intelligent enough to gauge safe interaction times.
  • Trade-off Between Attractiveness and Safety: There's a clear tension between making in-vehicle systems appealing to users (leading to willingness to communicate) and ensuring they don't compromise driving safety.

Key Results

  • Attitude: All drivers, regardless of age, preferred communicating with the young voice. Older adults found the young voice more affectionate, trustworthy, and less dominant than the old voice. Young drivers showed no significant difference in affection or trust between the voices.
  • Driving Performance (Accidents): All drivers had more accidents when using the young voice system compared to the old voice system or no voice system. Young drivers generally had more accidents than older drivers.
  • Driving Performance (Traffic Violations): Young drivers broke traffic regulations significantly more often than older drivers. The age of the voice had no effect on this measure.
  • Driving Performance (Swerving): Drivers without a voice system swerved the most. Drivers using the young voice system swerved significantly more than those using the older adult voice system. Young drivers also swerved more than older drivers.
  • Driving Time: Young drivers completed the course significantly faster than older drivers, indicating higher speeds. The age of the voice had no effect on driving time.
  • System Influence: Young drivers reported a negative influence of the young voice system on their driving, while older drivers reported no difference.

Limitations and Future Work

The authors acknowledge that the study used a driving simulator, and results may differ in real-world driving conditions. They also note that the conversational system was purely informational and initiated by the system, which might not reflect all types of in-vehicle voice interactions.

Future work should further investigate the detrimental effects of young voices on driving performance, especially for young drivers who are already a concern. The authors also suggest exploring why older drivers, despite preferring the young voice, might still be better focused on the driving task in certain scenarios. A critical area for future research is developing systems that can intelligently manage interactions, knowing when it is safe to communicate and when to cease interactions to prioritize driving safety.

Summary updated

BibTeX
@inproceedings{jonsson2009effects,
  author = {Ing-Marie Jonsson and Nils Dahlbäck},
  title = {{The effects of different voices for speech-based in-vehicle interfaces: impact of young and old voices on driving performance and attitude}},
  booktitle = {Interspeech},
  year = {2009},
}
Figure