The effects of low-pass filtering and random splicing on the perception of speech

In this paper the results are presented of an auditory description of 32 nonmanipulated, low-pass-filtered, and random-spliced speech samples. The description consisted of the ratings by three raters on 24 speech scales. The aim of this study was to establish the effects of the manipulations on the perception of speech. Insight into this question was gained by examining the reliability of, and the correlations between, the ratings in the three conditions, and by considering the outcomes of a number of t tests. It appeared that, after filtering, in addition to prosodic features, a number of voice quality characteristics remain present in the signal; articulatory information is eliminated. After splicing, not only voice quality features but also some articulatory and prosodic information remain present. Moreover, the study revealed that both content-masking techniques bring about systematic biases in the perception. These findings are relevant to emotion and personality research in which low-pass filtering and random splicing are used as masking techniques.