How do source-filter interactions reflects on glottal-flow parameters?

Résumé En

Linear source-filter theory has been successfully applied to voice analysis and synthesis for half a century (Fant, 1960). It is based on the strong assumption that source and filter can be modeled separately from each other, and do not interact. This is of course a first-order approximation of the physics of voice production. Several source-filter interaction effects may be observed on glottal flow, such as glottal pulse skewing and formant ripples (Childers and Wong, 1994). Glottal pulse skewing depends on vocal-tract acoustical loading, and it is one aspect of source-filter interaction which has been included in glottal flow models (speed quotient or asymmetry coefficient, Doval et al., 2006). Yet there is little quantitative knowledge about how glottal parameters may vary as a function of source-filter interaction. To explore further this question, a physically-based synthesizer has been used. It is based on a modified two-mass model of the vocal folds coupled to fluid flow description (Ruty et al., 2007). It includes a moving separating point of the airflow within glottal geometry. This model can produce oscillations (flow induced vibrations), and can be considered as a sound source. The produced acoustic wave propagates into a simple description of the vocal tract, which is based on linear acoustics. We can choose to take into account acoustical pressure to calculate transglottal pressure drop. Thus, the synthesizer can be run with (interactive) or without (non-interactive) acoustical feedback on the source. We present the variations in glottal-flow parameters between the interactive and noninteractive conditions. The parameters of interest here are the fundamental period, the open quotient and the asymmetry coefficient (equivalent to speed quotient). Fundamental period is calculated as the duration between two glottal closing instants detected on the synthesized glottal-flow pulse derivative. Glottal open time is measured as the duration between a glottal opening instant and the following glottal closing instant. Glottal opening time is measured as the duration between a glottal opening instant and the following instant of maximal glottal flow. Open quotient is derived as the ratio between glottal open time and fundamental period. Asymmetry coefficient is derived as the ratio between glottal opening time and glottal open time. For a fixed two-mass model configuration, the acoustical feedback is found to affect all glottal parameters, including the glottal fundamental frequency. As expected, modifications in vocaltract geometry only have an impact in the interactive condition. It affects mainly glottal-pulse skewing, but we can also notice an increase of the open quotient, and an increase of the fundamental frequency. References Childers, D. G. & Wong, C. F. (1994) Measuring and modeling vocal source-tract interaction., IEEE Trans Biomed Eng, vol. 41, n°7, pp. 663-671. Doval B., d'Alessandro C. and Henrich N. (2006) The spectrum of glottal flow models, Acta Acustica united with Acustica, vol. 92, pp. 1026-1046. Fant G. (1960) Acoustic theory of speech production. Mouton, La Hague N. Ruty, X. Pelorson, A. Van Hirtum, I. Lopez-Arteaga, A. Hirschberg. (2007) An in-vitro setup to test the relevance and the accuracy of low-order vocal folds models, The Journal of the Acoustical Society of America, vol. 121(1), pp. 479-490