Tree-based Clustering of Vowel Duration Ratio Toward Dictionary-based Automatic Assessment of Prosody in L2 English Word Utterances

Kohei Kitamura, Tsuneo Kato, Seiichi Yamamoto


Placing correct accents in producing a word is the first step for second language (L2) learners to acquire the rhythm of a language. To evaluate correctness of the contrast between long and short syllables, we have proposed a referential vowel duration ratio (R-VDR), which takes the ratio of the segmental duration between two vowels in consecutive syllables in reference to the magnitude relation of the duration between the same vowels in the same word uttered by native speakers. The R-VDR significantly improved the correlation between the objective and subjective assessment scores on prosody (subjective-objective score correlation). However, it requires a native speaker's reference utterance of the same word. To migrate from referencing native speakers' utterances to referencing a pronunciation dictionary, we applied tree-based clustering to the weights for computing the objective score. A preliminary experiment showed that rational clusters were formed by the resulting decision tree, although a weighted mean of log VDR with the clustered weights improved the subjective-objective score correlation slightly compared with the arithmetic mean of log VDR.


 DOI: 10.21437/SpeechProsody.2020-200

Cite as: Kitamura, K., Kato, T., Yamamoto, S. (2020) Tree-based Clustering of Vowel Duration Ratio Toward Dictionary-based Automatic Assessment of Prosody in L2 English Word Utterances. Proc. 10th International Conference on Speech Prosody 2020, 980-984, DOI: 10.21437/SpeechProsody.2020-200.


@inproceedings{Kitamura2020,
  author={Kohei Kitamura and Tsuneo Kato and Seiichi Yamamoto},
  title={{Tree-based Clustering of Vowel Duration Ratio Toward Dictionary-based Automatic Assessment of Prosody in L2 English Word Utterances}},
  year=2020,
  booktitle={Proc. 10th International Conference on Speech Prosody 2020},
  pages={980--984},
  doi={10.21437/SpeechProsody.2020-200},
  url={http://dx.doi.org/10.21437/SpeechProsody.2020-200}
}