It Sounds Like You Have a Cold! Testing Voice Features for the Interspeech 2017 Computational Paralinguistics Cold Challenge

Mark Huckvale, András Beke


This paper describes an evaluation of four different voice feature sets for detecting symptoms of the common cold in speech as part of the Interspeech 2017 Computational Paralinguistics Challenge. The challenge corpus consists of 630 speakers in three partitions, of which approximately one third had a “severe” cold at the time of recording. Success on the task is measured in terms of unweighted average recall of cold/not-cold classification from short extracts of the recordings. In this paper we review previous voice features used for studying changes in health and devise four basic types of features for evaluation: voice quality features, vowel spectra features, modulation spectra features, and spectral distribution features. The evaluation shows that each feature set provides some useful information to the task, with features from the modulation spectrogram being most effective. Feature-level fusion of the feature sets shows small performance improvements on the development test set. We discuss the results in terms of the most suitable features for detecting symptoms of cold and address issues arising from the design of the challenge.


 DOI: 10.21437/Interspeech.2017-1261

Cite as: Huckvale, M., Beke, A. (2017) It Sounds Like You Have a Cold! Testing Voice Features for the Interspeech 2017 Computational Paralinguistics Cold Challenge. Proc. Interspeech 2017, 3447-3451, DOI: 10.21437/Interspeech.2017-1261.


@inproceedings{Huckvale2017,
  author={Mark Huckvale and András Beke},
  title={It Sounds Like You Have a Cold! Testing Voice Features for the Interspeech 2017 Computational Paralinguistics Cold Challenge},
  year=2017,
  booktitle={Proc. Interspeech 2017},
  pages={3447--3451},
  doi={10.21437/Interspeech.2017-1261},
  url={http://dx.doi.org/10.21437/Interspeech.2017-1261}
}