05792293

a

05792293 Oehmen, Raoul Kirsner, Kim Fay, Nicolas Reliability of the manual segmentation of pauses in natural speech. Loftsson, Hrafn (ed.) et al., Advances in natural language processing. 7th international conference on NLP, IceTAL 2010, Reykjavik, Iceland, August 16--18, 2010. Berlin: Springer (ISBN 978-3-642-14769-2/pbk). Lecture Notes in Computer Science 6233. Lecture Notes in Artificial Intelligence, 263-268 (2010). 2010 Berlin: Springer EN

doi:10.1007/978-3-642-14770-8_30

Summary: Recent innovations regarding analysis of pauses in natural speech have necessitated the segmentation of increasingly small pause durations from the speech stream [1]. Identifying pauses and pause durations relies on human judgement. However the reliability of these judgements has yet to be established. This study investigated the reliability of multiple segmentations of four speech files. Results suggest that while inter-analyst reliability is moderate; intra-analyst reliability was high. Furthermore, inter-analyst variation appears to be related to the signal to noise ratio of the speech files. A further, analysis of the segmentation of one speech file demonstrated that a lack of reliability was associated with certain non-speech vocalizations, suggesting that reliability could potentially be increased with more precise guidelines for analysts.