The parallel data is a problem here — you can’t crowdsource ground truth because no one can record themselves with a different accent.
The parallel data is a problem here — you can’t crowdsource ground truth because no one can record themselves with a different accent.