Listeners are able to resolve subtle spectral-temporal details of speech waveforms when attempting to recognize words spoken in the listeners' native language. This may provide support for the existence of some processing mechanisms that are specific to the sounds of speech. A more parsimonious interpretation is that all highly familiar spectral-temporal patterns (speech and nonspeech) are processed more efficiently than less familiar sounds. A series of studies of word-length tonal patterns, conducted over the past 30 years, has provided some insights into the roles of uncertainty and familiarity in the discrimination and identification of speech and complex nonspeech sounds. Major conclusions from these studies are presented. The findings are generally consistent with the hypothesis that the "special" nature of speech is in large part due to its familiarity and its importance in conveying useful information.