> We prefer subtitles over machine-learning-based detection because the presence of a cough in subtitles means it was prominent enough for a person to write it down. Also our postdoctoral fellow was the only one who completed the Tensorflow tutorial.
lol