Modeling and perceiving of (un-)certainty in articulatory speech synthesis

This paper deals with the role of paralinguistic expression in articulatory speech synthesis. We describe two experiments which investigate the perception of certain vs. uncertain utterances produced by articulatory speech synthesis, using the system developed in [1]. Experiment 1 tests to what extent subjects are able to identify certainty and uncertainty as intended paralinguistic expressions in the acoustical signal by the varying acoustic cues intonation and delay. Further on, we investigate if (un)certainty influences the intelligibility of the synthetic utterances. Results show that the utterances are identified as intended with respect to (un)certainty. Regarding intelligibility, hardly any influence is measurable. Experiment 2 looks more in detail into the perception of uncertainty by using several levels. Therefore, not only intonation and delay are varied as acoustical cues but also fillers. Results show that our intended different levels of uncertainty indeed evoked different degrees of perceived uncertainty.