Exploring how Voice Assistants' vocal characteristics influence user engagement and decision-making in voice assistant interactions
Duration
Q2,3 2023, 1.5 months
My Role- Led research material design and experiment execution in collaboration with Computer Science and Merchandising teams
- Conducted online survey with 340 participants across multiple user groups
- Preprocessed data and performed comprehensive statistical and qualitative analysis of collected data
- Periodically presented findings and insights to both internal and external stakeholders
Methods
Figma,
Midjourney,
Maze,
After Effects
Introduction
When Voice Assistants like Alexa or Siri deliver an information, do their vocal characteristics such as tone, age, and gender influence users' trust, engagement, and decision making? As these assistants increasingly handle complex decisions, this research examines how vocal attributes affect user behavior, particularly in critical scenarios like online shopping.
Project Objective
Establish evidence-based guidelines for voice assistant design by investigating how the voice assistants' vocal tone affects users' decision-making in AI interactions, while balancing user engagement with voice assistant trustworthiness.
Research Question
Primary RQ: How does the perceived tone of a voice assistant's voice persuade participants and subsequently affect participants' purchase decisions?
Methodological Choices
Online Survey with Quantitative and Qualitative
Components Online survey enables standardized audio presentation across diverse geographical locations while maintaining ecological validity, as users naturally interact with voice assistants through personal devices in private settings. Quantitative measures capture persuasiveness and purchase intention with statistical precision, while qualitative open-ended responses provide explanatory context for understanding the mechanisms behind vocal tone effects.Alternative Methods
Lab experiment: Provides maximum experimental control and consistent audio equipment but creates artificial environments that reduce ecological validity and limit sample sizes.
Field Study: Offers highest ecological validity in natural usage contexts, but makes variable control extremely difficult with limited sample sizes.Why We Chose Online Survey
The research objective required measuring authentic persuasion effects in realistic voice assistant contexts. Online delivery preserved natural interaction environments while maintaining experimental rigor, which is critical since voice assistants are predominantly used in private, personal settings where trust and intimacy dynamics influence persuasion. The 340-participant sample size needed for robust statistical power was also feasible online, while technical concerns were mitigated through audio normalization and attention checks.
Sample Size (N=340)
Rationale: Power analysis determined N=105 per condition needed to detect medium effect sizes (Cohen's d=0.5) with 80% power at α=0.05. The achieved sample of 340 participants provided robust statistical power while remaining practically feasible for high-quality data collection.Between-Subjects Design Choice
Rationale: Between-subjects design prevented carryover effects where exposure to multiple vocal tones could create artificial contrast effects or demand characteristics. Voice characteristics form immediate impressions that persist across interactions, making within-subjects comparison problematic for measuring authentic persuasion responses in realistic contexts.
Participants were randomly assigned to one of three vocal tone conditions (positive, neutral, negative) to ensure balanced demographic distribution across groups. This randomization eliminated selection bias and enabled valid causal inferences about vocal tone effects onTone Validation with Pre-test
Rationale: Conducted stimulus validation study (N=78) to confirm that generated vocal tones were perceived as intended (positive, neutral, negative) by independent judges. This pre-test ensured experimental validity by verifying that acoustic manipulations successfully created the intended emotional perceptions before testing persuasion effects in the main study.
Research Timeline
Findings
User Insights
Impact
Academic Impact
Won Best Paper Award at CUI 2024, establishing foundational research for tone-specific conversational AI design. The work has been cited in subsequent voice interface research and industry design guidelines.
Industry Application
Findings directly inform voice assistant development at major tech companies, providing evidence-based recommendations for vocal tone optimization in commercial applications and voice commerce platforms.Design Guidelines
Created actionable design principles for conversational AI teams, balancing user engagement with ethical considerations and establishing standards for trustworthy voice interface development.