Synthesis of nonperiodic features of pathological voices
Brian C. Gabelman, Jody E. Kreiman, Bruce R. Gerratt, Abeer A. Alwan · The Journal of the Acoustical Society of America · 2001
The question has been raised as to how successfully measures of nonperiodic vocal energy such as spectrally shaped aspiration noise, jitter, and shimmer can be used in modeling pathological voices, and how well listeners distinguish between these quantities. This work studies how two such quantities, spectrally shaped aspiration noise and jitter, can be used to synthesize pathological voices. Automatic algorithms for analyzing and modeling these measures are described. Tokens of the vowel [a] were inverse filtered and analyzed using a cepstral-domain comb filter algorithm to extract the aperiodic component of the vocal source. The pitch trajectory was established with an interpolating tracker and was high-pass filtered and analyzed to measure jitter. Synthetic tokens were then computed using the calculated levels of source noise alone, jitter alone, and a ratio of both, and the results compared with the original token in listening experiments. The success of each method at modeling the original token across a variety of voice samples will be described, as will implications for speech synthesis and voice perception. [Work supported by NIH/NIDCD.]