Influences of Variations in Speech Intensity and Other Factors upon the Speech Spectrum
J. C. R. Licklider, Mones E. Hawley, Robert A. Walkling · The Journal of the Acoustical Society of America · 1955
In calculating speech intelligibility by methods such as French and Steinberg's, and in other phases of the designing of communication systems, one uses the long-time-average power density spectrum of speech. Several determinations of this spectrum have been made previously, but in each case the talkers talked at a fixed level. The present observations are aimed primarily to determine how the shape of the spectrum differs from “normal” when, as is necessary in some situations, the talker raises his voice considerably above normal level. Secondary questions concern the amount of variation among talkers and among speech samples and the length of sample required for stable measurement. Results, obtained with two different spectrum analyzers, are presented. As the speech level is increased, the spectrum changes noticeably—enough to affect the outcome of intelligibility calculations. The middle- and high-frequency components are raised much more than the low-frequency components. The spectrum of normal speech has its maximum ordinate at 100 or 200 cps, whereas the spectrum of very intense speech has its maximum ordinate at 700 or 800 cps.