Does Gender and Accent of Voice Matter?: An Interactive Voice Response (IVR) experiment

Abstract

We explore the impact of the gender and accent of the voice recording in Interactive Voice Response (IVR) systems in low-literate and patriarchal contexts. We conducted a small randomized control trial (RCT) with 62 participants, prompting them to identify myths and factual statements from a list of 10 prompts. One of four sets of recordings were randomized for each participant: Male formal (MF), Male Informal (MI), Female Formal (FF), and Female Informal (FI). We found that (a) male participants found male voices as providing more accurate information, (b) formal voice made myths seem accurate to male participants, (c) there was a significant impact of participant education level on correctly identifying myths, and (d) female participants were more knowledgeable about maternal health facts. Our study provides some basic guidelines on the potential characteristics of the voice used in IVR systems when deployment is in low-literate, and patriarchal communities.