Voice quality changes with age. In many cases, these voice changes result in a lower quality of life. Because one way of identifying these voice quality changes is through perceptually estimating talker age, correlations made between estimated talker age and acoustic analysis can provide insight to the possible physiological degeneration related to vocal function. While most perceptual studies investigating estimated talker age are cross-sectional, a longitudinal study of single speakers could provide additional details in the progressive degeneration of the voice quality. Nevertheless, one limitation of these studies is that perceptual ratings of voice quality or talker age in a longitudinal study could be biased by recording quality. Further, the spectral qualities of recordings from earlier decades are limited by the technology used. In this paper, a spectral-normalization filter was developed and applied to a corpus of recordings from an individual spanning about 50 years (1959 - 2007) to reduce this impact of these limitations. The filter was shown to be effective in normalizing the autospectra of the recordings and the fundamental frequency was unaffected by the filter. Preliminary subjective analysis suggests that the recording quality of all the files were perceptually similar.

This content is only available via PDF.