Speech acknowledgment systems have greater mistake rates for black individuals, study discovers
March 24, 2020 No Comments Tech Life Lisa Lawson

No, you’re not simply picturing it. Speech acknowledgment systems established by the similarity Amazon, Apple, Google, IBM, and Microsoft all have greater error rates when transcribing speech from black people than when doing so for white individuals. So discovers a study published today in the Procedures of the National Academy of Sciences, which notes that the increased occurrence of these systems in contemporary life risks preserving a new form of digital discrimination unless action is taken.”Automated speech acknowledgment(ASR)systems are now used in a range of applications

to transform spoken language to text, from virtual assistants, to closed captioning, to hands-free computing, “composed the study’s authors. “Our outcomes point to difficulties faced by African Americans in utilizing progressively prevalent tools driven by speech acknowledgment technology.”Researchers took 115 human-transcribed interviews– consisting of discussions with 73

black speakers and 42 white speakers– and compared them to the versions produced by the tech giant’s speech-recognition tools. They found an” average word mistake rate “of almost double(0.35 )when the system transcribed black speakers as compared to when it transcribed white speakers (0.19 ). “To put it simply, the systems worked visibly even worse for people of color. This is far from the first time we’ve seen evidence of bias ingrained in

the expected technology of tomorrow. In December of in 2015, a federal study as soon as again confirmed that facial-recognition tech is a biased mess. In October of 2017, we saw that Google’s text/sentiment-analysis tool showed signs of homophobia, bigotry, and antisemitism.

Nevertheless, it’s worth keeping in mind that with one quarter of U.S. grownups claiming to have at least one smart speaker in their houses, the bias discovered by today’s research study is most likely affecting tens of millions (if not more) individuals right now.

SEE ALSO: Federal research study confirms facial recognition is a prejudiced mess

“With adoption of speech recognition systems most likely to grow in time,” the research study authors continue, “we hope technology companies and other individuals in this field foreground the equitable advancement of these essential tools.”

That ‘d sure be nice, would not it?

About The Author

Leave a reply

Your email address will not be published. Required fields are marked *

%d bloggers like this: