Voice source variations in running speech: a study of Mandarin Chinese tones.

Number 951
Year 1995
Drawer 18
Entry Date 07/13/1998
Authors Löfqvist, Anders, Koenig, Laura L., and McGowan, Richard S.
Contact
Publication Vocal Fold Physiology: Voice Quality Control , edited by O. Fujimura and M. Hirano, Singular Publishing, Inc, San Diego, pp. 3-22.
url http://www.haskins.yale.edu/Reprints/HL0951.pdf
Abstract [Introduction] This paper examines variations in the sound source during naturally produced utterances by inverse filtering of the oral/nasal airflow. The relations between stress, tone, and intonation and the voice source have primarily been examined using acoustic analysis of fundamental frequency, and there exists a large body of data on F0 variations in speech. Aerodynamic studies of voice source in relation to changes in intensity and fundamental frequency have mostly been concerned with steady phonations. Early studies of airflow were restricted to average flow rates because of the limited bandwidth of the recording systems. Recording systems for airflow with a higher frequency response, such as the one described by Rothenberg (1973), in combination with inverse filtering procedures using digital signal processing techniques (cf. Javkin, Antonanzas-Barros, & Maddieson, 1987) have recently allowed more detailed analysis of voice source aerodynamics. For example, Holmberg, Hillman, and Perkell (1988, 1989) examined source pulse properties as a function of intensity and fundamental frequency. Such controlled studies are necessary for clarifying the basic relationships among the source aerodynamic parameters. However, because fundamental frequency and intensity often cover in speech, it is also important to examine these relationships in naturally produced utterances. A few studies have examined source properties in running speech using inverse filtering of the acoustic signal. Results presented by Gobl (1988) suggest that the excitation strength of the source pulse (taken as the magnitude of the derivative of glottal flow during the closing phase) is higher in stressed than in unstressed vowels. Pierrehumbert (1989) reports that the excitation strength varies with the fundamental frequency pattern of the utterance as well as with the overall voice level. Pierrehumbert and Talking (1992) show that the ratio of the fundamental to higher harmonics is affected by accent and by position relative to an intonation phrase boundary. The present experiment was designed to provide further information on voice source properties in naturally produced utterances. To obtain records of different F0 patterns in a controlled and systematic fashion, the tones of Mandarin Chinese were used as speech material. Mandarin is traditionally analyzed as having four lexical tones (cf.Howie, 1976; Tseng 1981; Xu, 1993). Thus, a syllable can have up to four different meanings, depending on the tone used with it. Because fundamental frequency contours serve to make linguistic distinctions in Mandarin, we reasoned the F0 variation in an utterance would be constrained, and that this would provide a way of eliciting controlled differences in pitch contour. The traditional descriptions of the Mandarin tones are as follows: tone 1 is described as high-level, tone 2 as mid-rising, tone 3 as low-rising, and tone 4 as high-falling. Figure1-1 shows the acoustic signal and fundamental frequency curves for single productions of each of the four tones, as produced by our female subject FW. These voice source data are interpreted in terms of the cover-body models of the vocal folds (Hirano, 1981; Fujimura, 1981; Titzel, Luschei, & Hirano, 1989; Titze, 1991; Farley, 1994). In a cover-body model, for a fixed subglottal pressure and state of glottal adduction, the factors that determine fundamental frequency-length and tension of the cover of the folds-are themselves determined by the activities of the cricothyorid muscle and the thyroarythenoid muscle. Muscle activation plots (e.g., Titze, 1991) show that the fundamental frequency does not uniquely specify the activities of these muscles. However, because muscle activations help to determine voice quality, as well as fundamental frequency, measures such as the open quotient (OQ) may help clarify the levels of muscle activation for a given fundamental frequency. Of course, source characteristics may not be determined solely by these intrinsic muscles of the larynx. Other influences may include activity of the sternothyroid (Niim, Horiguchi, & Kobayashi, 1991), sternohyoid (Sawashima & Hirose, 1983), and cricopharyngeus (Honda & Fujimura, 1991); the subglottal pressure; and the state of vocal fold adduction. The changes in some of these variables can be inferred from measures of pulse amplitude and time average flow. Thus, we generally should be able to infer more about laryngeal muscle activity with more measures of the glottal volume velocity than just fundamental frequency.
Notes There is no formal abstract.

Search Publications