DFKI-LT - A corpus based analysis of backchannel vocalizations
A corpus based analysis of backchannel vocalizations
Backchannel vocalizations play an important role in communicating listener intentions while the other person has the turn or other is talking. The communicative intentions behind backchannels not only transmit messages like 'I am listening' and 'I am with you', but also transmit listener affective states like excited, bored, confused, surprised, and so on. Synthesis of backchannel vocalizations is one of the focused research areas to improve emotionally colored conversational synthesis, and includes different research questions like where to synthesize, what to synthesize and what kind of acoustic properties have to be obeyed to communicate different affective states in different situations. Already a few attempts were made in this area of research; for example, the importance of affect bursts as a feedback in a conversation was investigated (Schröder et al, 2006) through listening tests, Nigel Ward and Wataru Tsukahara (2000) had developed some rules to generate backchannel responses in a conversation and investigated how to use low pitch regions as cues for backchannel responses. However, the analysis and identification of distinguishable types among backchannel vocalizations, their acoustic properties and affective states behind them have to be studied as they are crucial to improve interactive speech synthesis. This extended abstract explains about a method for collecting backchannel vocalizations and our ongoing work on annotation and a simple data and acoustic analysis of these vocalizations.
Files: BibTeX, pammi_schroeder2009a.pdf, pammi_schroeder2009a.pdf