Using prosodic information to predict when back-channels are appropriate in spontaneous dialogs has become somewhat of a reference problem for automatic discovery techniques. Here we present experiments with two ideas: the use of features derived from randomly generated pitch and energy filters, and the use of instance-based learning, specifically the Locally Weighted Linear Regression (LWLR) algorithm. For the task of predicting possible back-channel locations in Iraqi Arabic [6], we obtain 22% precision and 51% recall, which is as good as that obtained using a laboriously developed and hand-tuned rule.
Cite as: Solorio, T., Fuentes, O., Ward, N.G., Bayyari, Y.A. (2006) Prosodic feature generation for back-channel prediction. Proc. Interspeech 2006, paper 1724-Thu1FoP.11, doi: 10.21437/Interspeech.2006-601
@inproceedings{solorio06_interspeech,
author={Thamar Solorio and Olac Fuentes and Nigel G. Ward and Yaffa Al Bayyari},
title={{Prosodic feature generation for back-channel prediction}},
year=2006,
booktitle={Proc. Interspeech 2006},
pages={paper 1724-Thu1FoP.11},
doi={10.21437/Interspeech.2006-601}
}