ISCA Archive Interspeech 2021
ISCA Archive Interspeech 2021

Stacked Recurrent Neural Networks for Speech-Based Inference of Attachment Condition in School Age Children

Huda Alsofyani, Alessandro Vinciarelli

In Attachment Theory, children that have a positive perception of their parents are said to be secure, while the others are said to be insecure. Once adult, unless identified and supported early enough, insecure children have higher chances to experience major issues (e.g., suicidal tendencies and antisocial behavior). For this reason, this article proposes a speech-based automatic approach for the recognition of attachment in school-age children. The experiments are based on stacked RNNs and have involved 104 children of age between 5 and 9. The accuracy is up to 68.9% (F1 59.6%), meaning that the approach makes the right decision two times out of three, on average. To the best of our knowledge, this is the first work aimed at inferring attachment from speech in school-age children.


doi: 10.21437/Interspeech.2021-904

Cite as: Alsofyani, H., Vinciarelli, A. (2021) Stacked Recurrent Neural Networks for Speech-Based Inference of Attachment Condition in School Age Children. Proc. Interspeech 2021, 2491-2495, doi: 10.21437/Interspeech.2021-904

@inproceedings{alsofyani21_interspeech,
  author={Huda Alsofyani and Alessandro Vinciarelli},
  title={{Stacked Recurrent Neural Networks for Speech-Based Inference of Attachment Condition in School Age Children}},
  year=2021,
  booktitle={Proc. Interspeech 2021},
  pages={2491--2495},
  doi={10.21437/Interspeech.2021-904}
}