Automatic Geminate Insertion Algorithm for Japanese Audio Data

Generally, it is quite difficult for Japanese language learners to acquire special morae, namely, geminate, syllabic nasals and long vowels compared independent morae. Among these three geminate particularly difficult, takes much longer fully both production perception of it. Especially Chinese native speakers, previous studies has shown that are in terms the fact not only no found language, but also phonological interaction between accent tones. However, field speech acquisition, research making progress because a major problem, is, researchers themselves manually create acoustic experiment stimuli. Therefore, this study, as method solve we propose an algorithm automatically inserts into audio data used acquisition research. This automates insertion by performing processes order: mora extraction noise removal, matching original extracted mora, soundless duration geminate. The makes possible remove noise, which -50 dBFS continues 10ms or more, replace with instead, allowing speakers percept accuracy was equivalent result comparing modified phonology researcher generated algorithm. shows can be practical solution automation insertion.