Relevant Acoustic Group Features for Automatic Sleepiness Recognition

Dara Pir, Jarek Krajewski

Abstract

This paper compares the discriminating powers of various acoustic group features for the task of automatic sleepiness recognition using three different classifiers: Voted Perceptron, Simple Logistic, and Random Forest. Interspeech 2011 Sleepiness Sub-Challenge’s “Sleepy Language Corpus” (SLC) is used to generate the 4368 acoustic features of the official baseline feature set. The feature space is divided into Low-Level Descriptor (LLD) partitions. We consider the resulting feature space in groups rather than individually. A group feature corresponds to a set of one or more LLD partitions. The relevance of various group features to sleepiness state is then evaluated using the mentioned classifiers. Employing larger feature sets has been shown to increase the classification accuracy in sleepiness classification. Our results, however, demonstrate that a much smaller subset of the baseline feature set outperforms the official Sub-Challenge baseline on the SLC test data.

Download


Paper Citation


in Harvard Style

Pir D. and Krajewski J. (2018). Relevant Acoustic Group Features for Automatic Sleepiness Recognition.In Proceedings of the 4th International Conference on Information and Communication Technologies for Ageing Well and e-Health - Volume 1: ICT4AWE, ISBN 978-989-758-299-8, pages 209-214. DOI: 10.5220/0006779802090214


in Bibtex Style

@conference{ict4awe18,
author={Dara Pir and Jarek Krajewski},
title={Relevant Acoustic Group Features for Automatic Sleepiness Recognition},
booktitle={Proceedings of the 4th International Conference on Information and Communication Technologies for Ageing Well and e-Health - Volume 1: ICT4AWE,},
year={2018},
pages={209-214},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0006779802090214},
isbn={978-989-758-299-8},
}


in EndNote Style

TY - CONF

JO - Proceedings of the 4th International Conference on Information and Communication Technologies for Ageing Well and e-Health - Volume 1: ICT4AWE,
TI - Relevant Acoustic Group Features for Automatic Sleepiness Recognition
SN - 978-989-758-299-8
AU - Pir D.
AU - Krajewski J.
PY - 2018
SP - 209
EP - 214
DO - 10.5220/0006779802090214