||Estimation of glottal source waveform and vocal tract shape for singing-voice analysis
Takahashi, KyokoAkagi, Masato
2018 RISP International Workshop on Nonlinear Circuits, Communications and Signal Processing (NCSP2018)
694 , 2018-03-07 , Research Institute of Signal Processing, Japan
In this paper, an effective method to estimate the glottal source waveform and the vocal tract shape in singing voice was proposed based on ARX-LF model. Previous methods suffered from estimation of the glottal source waveform and the vocal tract shape in singing voices with high fundamental frequencies because of effects from forwarded periods. In the proposed method, parameters of the ARX-LF model were estimated accurately with exhaustive search in determined range and a simulated annealing method. Additionally, singing voice was re-synthesized using the estimated results of the vocal tract filter and periodic glottal source waveform with a length of settling time for considering the effects from forwarded periods. As a result of analysis using simulated singing voice data and actual sung voice data, the accuracy of estimation of the parameter values of the ARX-LF model from singing voices with wide range of fundamental frequency can be achieved by the proposed method.