In this paper, we investigate a DNN tone-based extended recognition network (ERN) approach to Mandarin tone recognition and tone mispronunciation detection. Given a toneless syllable sequence, a tone-based ERN is constructed by assigning five different tones to each toneless syllable, obtaining a fully expanded tonal syllable network. Next, Viterbi decoding is carried out on the tone-based ERN to find the best tone sequence. With respect to the tone recognition task, different acoustic units, and DNN configurations are compared. The experimental results show that tonal phone and longer DNN input window achieve better recognition performance. Moreover, we have applied confidence score extracted from tone-based ERN to verify whether L2 learners' tones are correctly pronounced. Compared with the conventional tone-based GOP (Goodness of Pronunciation) system, the proposed framework reduces the equal error rate by 10.98% relative.
|Titolo:||Using tone-based extended recognition network to detect non-native Mandarin tone mispronunciations|
|Data di pubblicazione:||2017|
|Appare nelle tipologie:||4.1 Contributo in Atti di convegno|