Christine H. Shadle
Dept. of Electron. and Comput. Sci., Univ. of Southhampton, Southhampton SO9 5NH, England
Much progress has been made towards establishing source and filter characteristics of fricative consonants by comparing the sound output by a human sustaining a fricative and by a mechanical model of that subject's static vocal tract shape. The problem can also be attacked by analysis of unsustained utterances, namely, repeated /pVFV/ sequences, using an analysis tool described elsewhere [Shadle et al., J. Phys. IV, Coll. C1 2, C1-295--C1-298 (1992)] that gives fine time resolution on the acoustic signal, and a variety of articulatory and aerodynamic data for the subjects. The repeated syllables form an ensemble, and ensemble averaging is performed across several tokens rather than the usual time-averaging within a single fricative. The resulting spectra show clear evidence of the expected changes in source (increase in high-frequency energy) and filter (format shift and pole-zero cancellation) during aVF transition. But unexpected changes are also apparent, e.g., evidence of a rounded vowel context enabling a different source strategy requiring less airflow in /pusu/compared to /pasa/. Phenomena such as devoicing in voiced fricatives can also be closely examined. Results for two subjects will be discussed. [Work supported in part by a collaborative EC SCIENCE award, CEC-SCI*0147C(EDB).]