ITSLanguage

ITSLanguage
Closed

ITSLanguage

Innovative speech technology for learning, teaching and assessment

The ITSlanguage project is a research project on language learning for Dutch learners of English. A large corpus of classroom recordings has been collected and used to develop our series of tools and models.

In our approach, a student’s attempt to pronounce a given phrase is compared directly against an audio example provided by their teacher.

Our proposed method produces a phoneme-level assessment, which is a far more challenging task than word, sentence or student level assessment. We introduce a binary error classification regime that allows an efficient pronunciation assessment, and which benefits from advanced acoustic modelling with deep neural networks (DNNs). The new method is also computationally efficient, as it uses DNN features and GMM-based binary classification rather than automatic speech recognition, as recent related work describes [7]. The cross-correlation of our best system and average human annotator reference scores is 0.72, with miss and false alarm rate around 19%. Automatic assessment is 81.6% correct with a high degree of confidence.

This work was undertaken in collaboration with ITSLanguage BV

Publications

2015

  • [PDF] [DOI] M. Nicolao, A. V. Beeston, and T. Hain, “Automatic Assessment of English Learner Pronunciation Using Discriminative Classifiers ,” in IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2015, Brisbane, Australia, 2015, pp. 5351-5355.
    [Bibtex]
    @inproceedings{nicolao_icassp2015,
    address = {Brisbane, Australia},
    author = {Nicolao, Mauro and Beeston, Amy V and Hain, Thomas},
    booktitle = {{IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2015}},
    doi = {10.1109/ICASSP.2015.7178993},
    month = {apr},
    pages = {5351--5355},
    project = {ITSLanguage},
    title = {{Automatic Assessment of English Learner Pronunciation Using Discriminative Classifiers }},
    year = {2015}
    }

Back to Top