In human speech, laughter has a special role as an important non-verbal element, signaling a general positive affect and cooperative intent. However, laughter occurrences may be categorized into several sub-groups, each having a slightly or significantly different role in human conversation. It means that, besides automatically locating laughter events in human speech, it would be beneficial if we could automatically categorize them as well. In this study, we focus on laughter events occurring in Hungarian spontaneous conversations. First we use the manually annotated occurrence time segments, and the task is to simply determine the correct laughter type via Deep Neural Networks (DNNs). Secondly we seek to localize the laughter events as well, for which we utilize Hidden Markov Models. Detecting different laughter types also poses a challenge to DNNs due to the low number of training examples for specific types, but this can be handled using the technique of probabilistic sampling during frame-level DNN training.
CITATION STYLE
Gosztolya, G., Beke, A., & Neuberger, T. (2019). Differentiating laughter types via hmm/dnn and probabilistic sampling. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11658 LNAI, pp. 122–132). Springer Verlag. https://doi.org/10.1007/978-3-030-26061-3_13
Mendeley helps you to discover research relevant for your work.