Second language (L2) speech is often annotated with the native phoneme categories. However, we often observe that an L2 speech segment generally deviates from a canonical phoneme and sometimes it is very difficult for linguists to annotate with any canonical phoneme label. We refer to these segments as non-native phonetic patterns. Existing approaches to mispronunciation detection and diagnosis (MDD) focus mainly on canonical mispronunciations, i.e. one canonical phoneme is substituted for another, aside from those deleted or inserted. To better represent L2 speech, this work explores non-native phonetic patterns (NN-PPs) of each native phoneme by an unsupervised approach. We apply an optimized k-means algorithm to cluster state-based phonemic posterior-grams, which are generated with a deep neural network. Then, to discover the NN-PPs related to each native phoneme, we perform forced alignment to divide L2 speech into segments grouped by native phonemes. We use the cluster sequences within segments derived from clustering results to represent different phonetic patterns of each native phoneme. Finally, we apply Cluster Sequence Analysis to discover each phoneme's potential NN-PPs. We verified experimentally that NN-PPs can extend the native phoneme categories to better describe L2 speech, which can enrich the existing approaches to MDD for better performance.
Cite as: Li, X., Mao, S., Wu, X., Li, K., Liu, X., Meng, H. (2018) Unsupervised Discovery of Non-native Phonetic Patterns in L2 English Speech for Mispronunciation Detection and Diagnosis. Proc. Interspeech 2018, 2554-2558, doi: 10.21437/Interspeech.2018-2027
@inproceedings{li18o_interspeech,
author={Xu Li and Shaoguang Mao and Xixin Wu and Kun Li and Xunying Liu and Helen Meng},
title={{Unsupervised Discovery of Non-native Phonetic Patterns in L2 English Speech for Mispronunciation Detection and Diagnosis}},
year=2018,
booktitle={Proc. Interspeech 2018},
pages={2554--2558},
doi={10.21437/Interspeech.2018-2027}
}