Melissa Epstein

This study examines the effects of prosody on voice quality variations in American English sentences. Research on voice quality variations in English requires both taking into account the various linguistic sources for changes in voice quality and adequately tracking and measuring the range of naturally occurring voice qualities in English. For instance, several consonants in English have glottal stop allophones that can cause glottalization of the preceding vowel, and there are personal and regional variations in voice quality. In this study, these effects are controlled by obtaining an idiosyncratic baseline value for each measurement for each word of the corpus. Furthermore, voice quality variations are tracked using both quantitative measurements (derived from the Liljencrants/Fant (LF) model of the glottal source) and a qualitative description of the waveforms to capture the range of voice qualities in the corpus. Modeled source pulses were evaluated to calculate the measurements OQ (open quotient), RK (glottal skew), EE (spectral intensity) and Linearity (spectral linearity).

Samples were taken from 3 speakers producing variants of a single sentence. Tokens included interrogative and declarative sentences, and sentences with phrase-initial and phrase-final prominent words. Results show that there are consistent effects of prominence and phrase boundaries on voice quality, both across speakers and among measurements. Both prominent words and phrase-initial words displayed a .tenser. voice quality than their non-prominent and phrase-final counterparts. A tense voice quality is associated in theory with greater compression of the vocal fold and greater force of closure of the arytenoids. Acoustically, tense voice quality is correlated with low values of open quotient and glottal skew, and high values of spectral intensity and spectral linearity. On the other hand, changes in tense or lax voice quality were not correlated with changes in pitch, nor were there any effects of pitch accent type or boundary tone on modal voice quality. There was, however, an increase in non-modal phonation associated with Low boundary tones, but not with low phonetic pitch in general.