The internal representation of vowel spectra investigated using behavioral response-triggered averaging.

Listeners presented with noise were asked to press a key whenever they heard the vowels [a] or [i:]. The noise had a random spectrum, with levels in 60 frequency bins changing every 0.5 s.