The current study investigated why the intelligibility of expressive speech in noise varies as a function of the emotion expressed (e.g., happiness being more intelligible than sadness), even though the signal-to-noise ratio is the same. We tested the straightforward proposal that the expression of some emotions affect speech intelligibility by shifting spectral energy above the energy profile of the noise masker. This was done by determining how the spectral profile of speech is affected by different emotional expressions using three different expressive speech databases. We then examined if these changes were correlated with scores produced by an objective intelligibility metric. We found a relatively consistent shift in spectral energy for different emotions across the databases and a high correlation between the extent of these changes and the objective intelligibility scores. Moreover, the pattern of intelligibility scores is consistent with human perception studies (although there was considerable individual variation). We suggest that the intelligibility of emotion speech in noise is simply related to its audibility as conditioned by the effect that the expression of emotion has on its spectral profile.
Cite as: Davis, C., Chong, C.S., Kim, J. (2017) The Effect of Spectral Profile on the Intelligibility of Emotional Speech in Noise. Proc. Interspeech 2017, 581-585, doi: 10.21437/Interspeech.2017-948
@inproceedings{davis17_interspeech,
author={Chris Davis and Chee Seng Chong and Jeesun Kim},
title={{The Effect of Spectral Profile on the Intelligibility of Emotional Speech in Noise}},
year=2017,
booktitle={Proc. Interspeech 2017},
pages={581--585},
doi={10.21437/Interspeech.2017-948}
}