The voice quality of pedagogical agent impacts learning and agent perceptions

Abstract Background The voices virtual on‐screen characters use has been shown to impact learning and perception outcomes. Recent replication research on these voices showed that synthetic voices were not a detriment if produced by a high‐quality engine with clear articulation. The current manuscript examines previous accent research that utilized now outdated engines, to determine if the impact of accents still holds with high‐quality engines and voice actors. Objectives To investigate the impact on learning and perceptions with pedagogical agents speaking in accented voices, synthetic voices, and the interaction between the two using modern voice engines. Methods This study is a between‐subjects two (accent) by two (type) factorial design to determine the impact the voice accent, voice type, and the interaction have on learning retention, learning transfer, mental effort efficiency, and perception measures. 197 participants were recruited from the online Amazon's Mechanical Turk with qualifications of 18 years of age, “normal or corrected‐to‐normal hearing”, and located with the continental United States of America. Results and Conclusions There were no significant differences between the accented conditions or interaction effects, deviating from previous research that showed impact of accents on learning. The synthetic condition had significantly lower knowledge retention, knowledge transfer, mental effort efficiency, and perception measures than the human professional. These findings demonstrate the importance of considering voice quality when designing pedagogical agents. Previous research showed synthetic voices perform as well as the average voice, and this research continues the narrative of voice quality by showing professional recordings outperform modern synthetic engines.