Abstract
Vocal effort is a major source of variability in speech processing. The present study examines its spectral effects from calibrated data recorded in 1977. The 97 talkers were instructed to vary their vocal effort in five degrees. Each sequence was represented by its sound level and its 1/3 octave long-term-average spectrum. After normalization to a common arbitrary level, comparing each spectrum to the others demonstrated that the original sound level could be recovered within a 5 dB error margin. A principal component analysis brought out several spectral features involved in the quantitative relationship between spectral shape and sound level.

This publication has 13 references indexed in Scilit: