Dynamic formant extraction of wa language based on adaptive variational mode decomposition

摘要

Wa language is one of Chinese minority languages spoken by the Wa nationality who lives in Yunnan Province, China. Until now, it has not been studied from the perspective of Engineering Phonetics. In this paper, for the above reason, by the adaptive variational mode decomposition (AVMD) we have investigated the dynamic formant characteristics of Wa language. Firstly, more precisely, use the synthetic dimension to split Wa language isolated words into voiceless and voiced segment, initials and finals. Secondly, use Linear Prediction Coding to estimate the first three formant frequencies and their bandwidths roughly. Thirdly, select the appropriate equilibrium constraint parameter and the number of decomposed layers so that Adaptive Variational Mode Decomposition (AVMD) can decompose the signal into some intrinsic mode functions (IMFs) without pattern aliasing. Fourthly, use the estimated formant frequencies and bandwidths to determine precisely the required IMFs. Fifthly, use the Hilbert transform to calculate the instantaneous frequency of the above determinate IMFs. Further, we implement the weight average operation on instantaneous frequencies to obtain the first three formant frequencies for each frame. Finally, comparing the first three formant frequencies obtained by the adaptive variance modal decomposition and by Praat software respectively, so we have drawn the conclusion that the relative correct rate of the former to the latter can reach 86% averagely in terms of the selected isolated words, which has shown that our method is effective on Wa language.