I find the contributions where the person doesn’t speak the recording at all or appears to be intentionally not reading the corpus to be more frustrating than those where the speaker is not perfectly fluent.