Comment so VR needs an Ebonics/English switch, BFD (Score 1) 155
The sample size is far too small and non-representative given that the speakers originated in only 5 geographical areas in the US -- Washington, DC, Rochester, Sacramento, Humboldt and Princeville, N.C. -- the almost exclusive AA community data source for the samples. (The southern inflection alone would account for almost all the error rate). Snipping up the limited original vocal samples into thousands of sound bites just to apply statistical metrics is hand waving at best. Akin to "We took DNA samples from every part of the body". To derive almost ANYTHING other than general cultural difference errors from this study is a serious stretch of logic.
This looks like another "To a hammer, everything looks like a nail' papers to fulfill the "Publish or Perish" requirement.