Towards Continuous Estimation of Dissatisfaction in Spoken Dialog

SigDial 2021

Nigel G. Ward, Jonathan E. Avila, Aaron M. Alarcon

Abstract: We collected a corpus of human-human task-oriented dialogs rich in dissatisfaction and built a model that used prosodic features to predict when the user was likely dissatisfied. For utterances this attained a F.25 score of 0.62, against a baseline of 0.39. Based on qualitative observations and failure analysis, we discuss directions for improving this result for practical utility.

paper

video

Audio Illustrations

• I think this is still too much
• no thank you
• thank you ... thank you so much
• my left front tire that is flat because of a nail
• Sample doomed dialog from the Appendix

Resources

• Corpus Description
• Corpus Data
• Code and Models
• weights of the model for predicting dissatisfied frames
• frame-wise correlations with the d (vs n) label

Nigel Ward's Publications