| Abstrakt: |
Eleven male subjects were required to read a text in both sober and alcohol intoxicated conditions. By means of statistical signal analysis, frequency distributions of fundamental frequency (F0), signal-to-noise ratio (SNR), ratio of first- to second-formant frequencies (F1/F2), variation speed of the frequencies F0, F1, F2, and the long-term average spectrum (LTAS) were determined. The distributions were examined for their suitability in discriminating between sober and intoxicated conditions. The SNR and F0 distributions as well as the LTAS discriminated with an error rate less than 5%. Combination of SNR and F0 profiles enabled correct discrimination in all cases. The parameter F1/F2 describing the articulation varied strongly among individuals. It was modified only with high levels of blood alcohol. Frequency variation speeds were not altered by intoxication. Speaker recognition by means of LTAS was more difficult in the intoxicated condition. The principal effect of intoxication on speech was interpreted as a perturbation of laryngeal movement control, where long-term voice effort was found to produce similar effects to alcohol intoxication. On the basis of the present results and various other factors (ambiguity of the sources of the acoustic effects, expense of the procedure), application of acoustic analysis in forensic medicine for recognition of low-level alcohol intoxication is considered inexpedient. [ABSTRACT FROM AUTHOR] |
