Akustische und perzeptive Analysen von Sprache unter Alkoholeinfluss

In this thesis, several acoustic and perceptive studies on the
influence of alcoholic intoxication on speech are presented. The
speech samples for the experiments are taken from the Alcohol
Language Corpus, a large database containing recordings of 162 (77
female and 85 male) speakers in both sober and intoxicated
conditions with various blood alcohol concentrations. An acoustic
analysis of recordings of 148 speakers in intoxicated and sober
condition reveals a significant increase of the mean and also of
the interquartile ranges of the fundamental frequency in the
majority of speakers. A large inter-speaker variability was found,
but no correlation with their individual blood alcohol
concentration. An analysis of the microperturbation parameters
jitter and shimmer in the vowels /a:/ and /i:/ showed no consistent
results and a large variance among the data. Jitter is
significantly lower in intoxicated condition but only in the vowel
/a:/. Shimmer is found to be significantly higher in intoxicated
condition but only for male speakers in the vowel /a:/ and in read
speech of male speakers in the vowel /i:/. Additionally, a forced
choice discrimination test was conducted with speech samples of 161
speakers. 71 naive listeners reached an average discrimination rate
of 63,1% which is well above chance. No gender-dependent effects
could be found. The ability of the listeners to pick the correct
stimulus was higher for speakers with high blood alcohol
concentrations, with no connection to the drinking habits of the
speakers. To see, whether fundamental frequency also functions as a
perceptual cue to reveal a person’s intoxication solely by the
speech signal, the results of the perception test were compared
with the acoustic measurements of fundamental frequency of the
speakers. Those who tend to use higher fundamental frequency in
intoxicated condition were judged correctly more often. Similar
results could be found for speakers who use larger pitch ranges.
The listeners also showed a general preference for the stimulus
with higher fundamental frequency and higher pitch range. To
further test fundamental frequency as a perceptual cue, two
perception tests with manipulated stimuli were conducted. In the
first one, the fundamental frequency effects in the stimulus in
intoxicated condition were compensated by adjusting mean
fundamental frequency and the interquartile range of the
intoxicated stimulus to that of the sober stimulus in each
discrimination pair. In the second one, fundamental frequency
effects were simulated in a sober stimulus and compared to another
sober stimulus in one pair. The simulation of fundamental frequency
effects was performed by up-shifting and stretching the contour by
5%, according to the findings in the acoustic study. In the test
with compensated fundamental frequency effects, listeners did not
perform worse than in the basis test. In the second test with
simulated fundamental frequency effects, they show a slight
tendency to pick the manipulated stimulus as the intoxicated one.
Fundamental frequency seems not to function as a strong perceptual
cue for listeners in revealing a person’s intoxication, maybe
changes of fundamental frequency occur as a side-effect of other
effects that play the major role. These could be due to other
acoustic features and/or maybe linguistic or paralinguistic
information. Listeners may use fundamental frequency as kind of a
fall-back feature if no other signs of intoxication are present as
is the case in the test with simulated effects. Probably listeners
do not rely on fundamental frequency as a perceptual cue, because
it is also prone to changes caused by other speaker states.