The role of voiced consonant duration in sung vowel-consonant and consonant-vowel recognition

Sung text intelligibility is often a problem, especially in reverberant acoustics, at high pitch, and in the presence of a loud accompaniment. This study aims to discover whether elongating the duration of voiced consonants /m/, /n/, /l/, and /v/ in sung vowel-consonant (VC) and consonant-vowel (CV) sequences improves their recognition. Perception tests were conducted with 42 participants, and the data were analyzed using generalized linear mixed models. Results showed that consonant durations of 20-35 ms are sufficient for a near-perfect recognition when singing at close-to-speaking pitch in non-reverberant acoustics and without accompaniment. However, in reverberant acoustics for VC sequences, longer consonant durations allow the reverberation from the preceding vowel to fade more fully during the consonant, reducing masking by the room reverberation and thereby improving recognition. Also, elongating consonant duration up to about 200 ms (or even longer) improved consonant recognition in the case of stimuli with added Brown Noise, imitating orchestral accompaniment, whereas only negligible impact on the order of succession (CV versus VC) was observed. Recognition tended to be poorer at higher pitch and with longer reverberation.