Many speech sounds, such as vowels, exhibit a characteristic pattern of spectral peaks, referred to as formants, the frequency positions of which depend both on the phonological identity of the sound (e.g. vowel type) and on the vocal-tract length of the speaker. This study investigates the processing of formant information relating to vowel type and vocal-tract length in human auditory cortex by measuring electroencephalographic (EEG) responses to synthetic unvoiced vowels and spectrally matched noises. The results revealed specific sensitivity to vowel formant information in both anterior (planum polare) and posterior (planum temporale) regions of auditory cortex. The vowel-specific responses in these two areas appeared to have different temporal dynamics; the anterior source produced a sustained response for as long as the incoming sound was a vowel, whereas the posterior source responded transiently when the sound changed from a noise to a vowel, or when there was a change in vowel type. Moreover, the posterior source appeared to be largely invariant to changes in vocal-tract length. The current findings indicate that the initial extraction of vowel type from formant information is complete by the level of non-primary auditory cortex, suggesting that speech-specific processing may involve primary auditory cortex, or even subcortical structures. This challenges the view that specific sensitivity to speech emerges only beyond unimodal auditory cortex.