Voice assistant technology continues to underperform on children's speech.

Area of Science:

• Human-Computer Interaction
• Developmental Psychology
• Speech Recognition Technology

Background:

• Voice assistant (VA) technology is becoming ubiquitous in daily life, including for young children.
• Limited research exists on the performance of VA systems in understanding the speech of children under five years old.

Purpose of the Study:

• To evaluate the accuracy of voice assistant (VA) systems (Siri and Alexa) in transcribing speech from 2-, 3-, and 5-year-old children.
• To compare VA performance with human listeners (mothers and undergraduates) in understanding children's utterances.

Main Methods:

• Tested two versions of Siri and one version of Alexa on their ability to transcribe utterances from children aged 2, 3, and 5 years.
• Human listeners, comprising mothers and undergraduate students, also transcribed the same child utterances for comparison.

Main Results:

• While recent improvements in Siri's performance were noted, both Siri and Alexa models demonstrated significant challenges in accurately transcribing children's speech.
• Human listeners consistently outperformed VA systems across all age groups, showing a marked advantage with younger children's speech.
• Performance disparities were most pronounced for the youngest participants (2-year-olds).

Conclusions:

• Current voice assistant (VA) technologies, including leading models like Siri and Alexa, exhibit limitations in understanding the speech patterns of children under five.
• Human auditory perception remains superior to current VA technology for interpreting children's speech, highlighting a critical area for technological advancement.
• Further research and development are needed to improve the inclusivity and effectiveness of VA technology for young users.