Improve low-resource non-native mispronunciation detection with native speech by articulatory-based tandem feature

In this paper, we propose a method to improve detecting the mispronunciation type of the non-native learners. In order to cope with the low-resource condition of non-native speech and the difference of native and non-native speech, the following efforts are made: 1) train acoustic model with the low-resource non-native data; 2) introduce the articulatory-based tandem feature; 3) pool auxiliary native data and non-native data together to train the articulatory-based MLP system. We take Chinese learning English for example, and select 1h speech to imitate the low-resource non-native speech situation. In addition, it's studied the combination of pitch and different articulatory-based tandem feature with different input feature (PLP, MFCC, Fliterbank). The experiments show that the proposed method improves the performance obviously. The phone recognition accuracy is improved by 2.99% and the mispronunciation type accuracy is improved by 2.27%.