Quantifying the perceptual value of lexical and non-lexical channels in speech