Formant-based representation of auditory similarity among vowel-like sounds.

Can the acoustic properties of vowels, in particular, formant frequencies, and formant amplitudes, be scaled so that the auditory relations among vowels are representable by their positions in a formant frequency space? Experiments are described in which carefully selected sets of five-formant vowel-like sounds were presented to listeners. Auditory judgements were elicited sufficient to evaluate the conformity of steady state vowel perception with various algebraic forms. Test results demonstrate that if intervowel distances are organized on a formant-by-formant basis, as for example in the commonly used Euclidean distance metric, no scaling of listeners' responses using a formant-based distance, two factors are required for each formant: Scaled formant amplitude levels must be included in the representation, and the difference of the scaled formant frequencies must be weighted by the formant amplitude levels.