“…Most research relating to using prosody for audio search has focused on detecting dialog activities that people might like to search for. Prosody-based classifiers can, for example, spot interactional "hotspots" where the speakers are unusually involved (Wrede and Shriberg, 2003;Oertel et al, 2011), conflicts (Kim et al, 2012), agreements on action items (Purver et al, 2007), various emotional and attitudinal states and stances (Toivanen and Seppänen, 2002;Wollmer et al, 2013), and dialog acts such as question, apology, promise, and persuasion attempt (Larson et al, 2011;Freedman et al, 2011). This work has shown many dialog activities are indeed associated with characteristic prosodic features and patterns.…”