Singers convey meaning via both text and music. As sopranos balance tone quality and diction, vowel intelligibility is often compromised at high pitches. This study examines how sopranos modify their vowels against an increasing fundamental, and in turn how such vowel modification affects vowel intelligibility. We examine the vowel modification process of three non-central vowels in Cantonese ([a], [ɛ] and [ɔ]) using the spectral centroid. Acoustic results suggest that overall vowel modification is conditioned by vowel height in mid-ranges and by vowel frontness in higher ranges. In a following perception task, listeners identified and discriminated vowels at pitches spanning an octave from A4 (nominally 440 Hz) to G♯5 (nominally 831 Hz). Results showed that perceptual accuracy rates of the three vowels’ match their acoustic patterns. The overall results suggest that vowels are not modified in a unified way in sopranos’ voices, implying that research on sopranos’ singing strategies should consider vocalic differences.