Mandarin Chinese Consonants Landmark Detection Based on Segmented Phoneme Classification

For the acoustic landmarks of Mandarin Chinese, the acoustic landmarks based on phonetics knowledge annotation and data-driven (such as CTC mechanism and attention mechanism) have good consistency. Among them, there are five Chinese consonants whose acoustic landmarks are quite different through the two detection mechanisms. In order to determine whether the Chinese acoustic landmarks of the five disputed phonemes are in the segment of initial or final, this study will take the Chinese corpus of Chinese native speakers as the research object, adopt three different segmentation principles to segment the phoneme samples, and use the segmented phoneme unit to classify the phonemes. The part with higher accuracy is regarded as the area where the acoustic landmarks exist, so as to locate the acoustic landmark of Chinese fricative. The experimental results show that the acoustic landmarks of /j/, /sh/, /x/ and /zh/ are in finial, while the acoustic landmarks of /k/ are in transit.