Analyzing Algorithmic Bias in Voice Recognition Technologies
See our final paper for a longer report of our findings.
You can also explore visualizations + summaries in our presentation.
“Everyone who speaks a language, speaks it with an accent.”
This project seeks to analyze algorithmic bias via regional accents in voice recognition technologies built by four prominent technology companies: Google, Amazon, Microsoft, and IBM. Given the increasing use of voice recognition technology in today’s society, it is important to assess claims to universal accessibility. Thus, this project explores the accuracy of four speech-to-text technologies with respect to English spoken by individuals with a variety of common international accents.
Our analysis finds that IBM is the top-performing technology, while Microsoft’s Bing speech-to-text consistently performs the worst. Additionally, we find clear differences in the accuracy of these technologies by accent, with three of the four technologies performing considerably better accuracy-wise on English spoken with a US American accent than on any other accent. Our analysis also finds particularly troublesome implications for use for those speaking English with Vietnamese or Spanish accents, as all four technologies perform poorly in these categories. These discrepancies have significant implications for the accessibility of hands-free and voice recognition technologies for individuals speaking English with a non-US American accent.