Acoustic characterization of the glides /j/ and /w/ in American English

Acoustic analyses were conducted to identify the characteristics that differentiate the glides /j,w/ from adjacent vowels. These analyses were performed on a recorded database of intervocalic glides, produced naturally by two male and two female speakers in controlled vocalic and prosodic contexts. Glides were found to differ significantly from adjacent vowels through RMS amplitude reduction, first formant frequency reduction, open quotient increase, harmonics-to-noise ratio reduction, and fundamental frequency reduction. The acoustic data suggest that glides differ from their cognate high vowels /i,u/ in that the glides are produced with a greater degree of constriction in the vocal tract. The narrower constriction causes an increase in oral pressure, which produces aerodynamic effects on the glottal voicing source. This interaction between the vocal tract filter and its excitation source results in skewing of the glottal waveform, increasing its open quotient and decreasing the amplitude of voicing. A listening experiment with synthetic tokens was performed to isolate and compare the perceptual salience of acoustic cues to the glottal source effects of glides and to the vocal tract configuration itself. Voicing amplitude (representing source effects) and first formant frequency (representing filter configuration) were manipulated in cooperating and conflicting patterns to create percepts of /V#V/ or /V#GV/ sequences, where Vs were high vowels and Gs were their cognate glides. In the responses of ten naïve subjects, voicing amplitude had a greater effect on the detection of glides than first formant frequency, suggesting that glottal source effects are more important to the distinction between glides and high vowels. The results of the acoustic and perceptual studies provide evidence for an articulatory-acoustic mapping defining the glide category. It is suggested that glides are differentiated from high vowels and fricatives by articulatory-acoustic boundaries related to the aerodynamic consequences of different degrees of vocal tract constriction. The supraglottal constriction target for glides is sufficiently narrow to produce a non-vocalic oral pressure drop, but not sufficiently narrow to produce a significant frication noise source. This mapping is consistent with the theory that articulator-free features are defined by aeromechanical interactions. Implications for phonological classification systems and speech technology applications are discussed. Thesis Supervisor: Kenneth N. Stevens Title: Emeritus Professor of Electrical Engineering and Health Sciences and Technology