| Effects of the common cold impair the performance of automatic speaker recognition systems. This research examines cold-related anomalies in speaker recognition by analyzing changes in both acoustic phonetic and speaker recognition parameters. Ten adult male and 10 adult female talkers were recorded during a cold, and again once normal health returned. Formants (F1, F2), nasality variables (A1, P1n, A1-P1n), and cepstra (c2) were chosen as experimental parameters for this study due to the theoretical effect of a cold-related nasal obstruction on each measurement. All variables (except P1n) measured during the cold sessions were hypothesized to have lower values than those same variables measured during the healthy session. Doubly multivariate ANOVA (repeated measures) statistical tests revealed significant differences in formant frequencies for F1 and F2 for the cold and healthy conditions (with-in subject). The frequencies for both F1 and F2 were lower for the cold condition than for the healthy condition (F1cold < F1 healthy; F2cold < F2 healthy). Differences in frequency for F1 and F2 were also evident between male and female talkers (between subject). The amplitude of nasality variable P1n, a spectral peak observed in nasal consonants near 1 KHz, also showed significant differences between cold and healthy conditions and between male and female speakers. The second mel-frequency cepstral coefficient, c2, showed differences between cold and healthy states, but the results were not significant. Intra-speaker variability was observed in c2, and there were significant interactions for /ng/ and /n/. Integrating literature in linguistics, medicine, and electrical engineering was an intentional multidisciplinary approach. Information from these fields was used to identify shifts in formant frequencies of nasalized vowels, reveal the presence of substantial changes within the ethmoid sinuses of the nasal cavities during a cold, and identify cepstral parameters as the primary feature used in automatic speaker recognition systems. Overall, this research contributes to the communications sciences and disorders discipline by providing analyses of acoustic and cepstral parameters of "cold-speech" that can be added to the pool of research required to improve the developmental framework of automatic speaker recognition systems and other state-of-the-art speech technologies. |
