In this paper, we describe a new database with audio recordings of non-native (L2) speakers of English, and the perceptual evaluation experiment conducted with native English speakers for assessing the prosody of each recording. These annotations are then used to compute the gold standard using different methods, and a series of regression experiments is conducted to evaluate their impact on the performance of a regression model predicting the degree of naturalness of L2 speech. Further, we compare the relevance of different feature groups modelling prosody in general (without speech tempo), speech rate and pauses modelling speech tempo (fluency), voice quality, and a variety of spectral features. We also discuss the impact of various fusion strategies on performance.Overall, our results demonstrate that the prosody of non-native speakers of English as L2 can be reliably assessed using supra-segmental audio features; prosodic features seem to be the most important ones.

This paper deals with databases that combine different aspects: children's speech, emotional speech, human-robot communication, cross-linguistics, and read vs. spontaneous speech: in a Wizard-of-Oz scenario, German and English children had to instruct Sony's AIBO robot to fulfil specific tasks. In one experimental condition, strictly parallel for German and English, the AIBO behaved `disobedient' by following it's own script irrespective of the child's commands. By that, reactions of different children to the same sequence of AIBO's actions could be obtained. In addition, both the German and the English children were recorded reading texts. The data are transliterated orthographically; emotional user states and some other phenomena will be annotated. We report preliminary word recognition rates and classification results.