Topic change detection based on prosodic cues in unimodal setting

Speech is considered as the most dominant way for people to interact with each other. But speech cannot be merely characterized as sequence of sound units. There are some non-verbal elements, such as prosodic cues, that lend naturalness to speech. Prosodic cues such as rhythm, stress and intonation are crucial to investigate because they play an important role in human-human as well as in human-machine communication. They help to compensate many hidden meanings omitted from spoken language. The goal of this study was to investigate the correlation between non-verbal acoustic cues and topic shift in unimodal setting and to show how prosodic information can be used as a signal to detect topic change within conversation.