Text this: Evaluating Low-Level Speech Features Against Human Perceptual Data