Computer measurements of breathy voice quality
Peter Ladefoged, Norma Antonanzas · The Journal of the Acoustical Society of America · 1984
Most measures of phonation types analyze samples of steady-state vowels that last several seconds. Such segments do not occur in natural languages. The principle used in the current measure is that within a single period of a voiced speech sound, a waveform with no breathiness should be the same as the waveform produced by the previous glottal pulse. But when a sound is produced with a turbulent glottal flow there will be randomness. Hence an appropriate measure of breathiness is the variance of the difference between the points in a window with a length slightly less than one period and the points in the most similar window of equal length in one of the immediately following periods. The measure can be used on segments of speech lasting only about 150 ms. It is largely unaffected by the changes in glottal pulse frequency that correspond to the intonation, and by period to period jitter. The measure has been tested on two languages that use phonological contrasts dependent on the difference between breathy voice and modal (regular) voice. [Work supported by NINCDS.]