The long-term average spectrum as a measure of voice stability

This paper aims at providing a methodological answer to the topic of voice stability. Seventeen experimental subjects read a standard text 5 times during a 2-week period. The data were obtained through the Long-Term Average Spectrum and were analyzed with a twofold procedure: (i) analysis of the absolute energy values at different frequency points throughout the length of the spectrum, and (ii) analysis of the relative values obtained by subtraction of each of two consecutive frequency values analyzed. The results obtained with the first procedure indicate that the differences that exist between the various sessions are centered mainly in the frequencies below 0.6 kHz and in the area of 4 kHz, following a linear tendency. The fact that the differences between sessions disappear when employing relative measures may indicate that the utilization of these measures eliminates the sources of systematic or aleatoric error can be introduced during a recording or in the period of time between two consecutive recording sessions.