Periodicity, spectral and electroglottographic analyses of pressed voice in expressive speech

Abstract

Pressed voice is a type of voice quality produced by pressing/straining the vocal folds, which often appears in Japanese conversational speech when expressing paralinguistic information related to emotional or attitudinal behaviors of the speaker. With the aim of clarifying the acoustic and physiological features involved in pressed voice production, we conducted periodicity, spectral and electroglottographic (EGG) analyses on pressed voice segments extracted from spontaneous dialogue speech of several speakers. Periodicity analysis first indicated that pressed voice is usually accompanied by creaky or harsh voices, having irregularities in periodicity, but can also be accompanied by periodic voices with fundamental frequencies in the range of modal phonation. Spectral analysis indicated power is usually reduced in low frequency components of pressed segments. A spectral measure H1'-A1' was then proposed for characterizing pressed voice segments which commonly has few or no harmonicity. H1'-A1' was shown to be effective for identifying most pressed segments, but fails when nasalization occurs. Vocal fold vibratory pattern analysis from the EGG signals revealed that most pressed voice segments (including nasalized vowels) are characterized by glottal pulses with closed intervals longer than open intervals on average, regardless of periodicity.