Detection of Prominent Words in Oral Reading by Children

Kamini Sabu, Preeti Rao


The evaluation of oral reading skills is considered an important component of language education in school. Compared with word decoding skill, prosodic fluency typically takes children much longer to achieve. Prosodic fluency, however, is linked to comprehension making its evaluation very useful in an automatic reading assessment system. We consider the detection of prominent words in recordings of oral reading by children, who display good word recognition but varying degrees of prosodic fluency. The manual annotation of prominent words proves to be relatively challenging, likely due to inconsistencies by our speakers with respect to top-down lexical cues. Acoustic-prosodic features drawn from prominence detection research on adult speech are tested on the annotated data using a random forest classifier. Normalized maximum syllable duration and F0 (fundamental frequency) derived features turn out to be important predictors of word prominence with their relative importances being highly speaker dependent.


 DOI: 10.21437/SpeechProsody.2018-64

Cite as: Sabu, K., Rao, P. (2018) Detection of Prominent Words in Oral Reading by Children. Proc. 9th International Conference on Speech Prosody 2018, 314-318, DOI: 10.21437/SpeechProsody.2018-64.


@inproceedings{Sabu2018,
  author={Kamini Sabu and Preeti Rao},
  title={Detection of Prominent Words in Oral Reading by Children},
  year=2018,
  booktitle={Proc. 9th International Conference on Speech Prosody 2018},
  pages={314--318},
  doi={10.21437/SpeechProsody.2018-64},
  url={http://dx.doi.org/10.21437/SpeechProsody.2018-64}
}