Voice Recognition Tech from Major Tech Giants Shows Racial Bias, Study Finds

According to the New York Times, Google, Amazon, Apple, Microsoft and IBM as well as their voice recognition technologies, are on the dock. For researchers, the technology used by these digital giants would have misidentified 35% of the words spoken by people of color.

Some attribute this error to the performance of voice recognition tools. If the voice recognition systems offered by Apple, Amazon, Microsoft, IBM and Google make very few mistakes when it comes to understanding the words spoken by white people, when these people are of color, the error rate is much higher.

According to the study originally published in the Proceedings of the National Academy of Sciences, the technologies qualified as incomprehensible only 2% of the audio excerpts of white people, and this multiplied by 10 to reach 20% in the case of audio of black people.

Other studies also reveal that voice recognition, in addition to being racist, would also be sexist. For the Stanford study, one of the explanations for the racism and sexism of voice recognition put forward, would be that the voice recognition systems would be defective and not tested enough by the companies that create them. These would train the technologies on databases lacking in diversity, it is argued.