Examination of the stress effect on speech quality

Corinna Kann and Verena Pyschny

The poster presentation is about the building and the comparison of a corpus which is divided into a stressed and an unstressed part. The three speakers we took for our small speech data bank were students aged 22 to 24. They had to explain a simple picture of a horse but were only allowed to use geometric terms in their explanations. Another person, who was informed about the course of the test, had to paint a nearly perfect copy of the original picture with the help of the speaker's explanation. The painting person knew how the picture had to look like and so the painter was able to manipulate the speaker: In the first part of the test, the painter cooperated with the speaker and painted a nearly perfect copy. In the second part of the test, however, a second painter had to understand the speaker's explanations and this painter pretended to be a bit slow on the uptake. Because of this fact and because of the experiment conductor's pleading to explain not as long-winded as in the first part the speaker was stressed. We examined the results of the speech material with regard to speech rate, elisions, basic frequency behaviour, volume, self-interruptions and vowel lengths. The hypotheses that the speech in the stressed part would be faster, more mumbled, higher, louder, more often self-interrupted and that the vowel length would be shorter was often confirmed. But we also found a lot of  inconsistency to our main hypotheses. This fact seems to be quite normal, because a speech data bank which is built of the speech material of only  three speakers isn't very representative and general.