Skip to content

There Is a Racial Divide in Speech-Recognition Systems, Researchers Say

nytimes.com
5 pointsfvrghl2 comments
On HN

Comments

Ugh--this is not a racial problem.

Lots of people of all different races--including white--use pronunciation and vernacular that differs significantly from the typical Midwestern speaker. I interact regularly with Caucasians that I can barely understand myself, let alone asking a typical commercial speech recognition system. On the flip side, quite a few black people speak in a manner that is indistinguishable from other races.

Universal recognition is the real problem to be solved. We want everyone to be understood.

True, and that is important, but there certainly is a correlation. Good luck writing speech recognition software to deal with this:

https://ifunny.co/video/damn-wtf-we-really-talk-like-that-63...

We may want everyone to be understood, but this goes far beyond speech recognition software. Humans often can't understand humans. There are lots of people who really don't talk well at all.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.