RSS

Simultaneous talk in conversation

Investigator: Emina Kurtic Supervisor: Guy J Brown

Traditional psycholinguistic research has led to major advances in the understanding of speech processing. However, the gap between the manipulated stimuli of laboratory experiments and real conversational speech shows up when attempts are made to apply findings to real word situations. To bridge the gap, this study proposes a novel methodology that works on naturally occurring conversational speech, combining signal processing with conversation analysis (CA).

While systematic social practices ensure that speakers talk one at a time, on occasion these practices break down, leading to simultaneous talk. Such occasions offer insight into on-line processing abilities, as speakers attempt to resolve the social interactional problem posed by finding themselves in overlap. There have been proposals in the CA literature, based on subjective listening, as to the various phonetic modifications that occur in different types of overlap, e.g. interruptions, accidental simultaneous start-ups. Recently, computational algorithms for the separation of simultaneous speech have been devised, which offer the possibility of an objective analysis of these modifications.

The project will employ audio and video recordings of naturalistic interactions. CA analysis of overlapping talk will be carried out. Separation of speech from overlapping speakers will be investigated using cues such as fundamental frequency. The separated voices will be analysed in terms of intonation, timing etc., for an objective characterisation of different types of overlap sequence. The project will also suggest objective measures for the detection of overlapping speech in audio recordings, which is a key problem in the automatic transcription of recorded meetings.