Analysis of nasals and nasalized vowels based on branched tube models

The coupling of the nasal and mouth cavity at the velum is essential for the production of nasals and nasalized vowels. Therefore branched tube models are discussed and an algorithm is proposed to estimate the model parameters from speech signals. The tube model is realized by lattice structures in discrete time. The estimation algorithm is processed in two steps. The zeros are estimated starting with a general ARMA-estimation determining the parameters of the coupled tube branch. Hereafter the remaining model parameters are estimated by the minimization of the output power of the entire inverse filter, which is carried out iteratively. The analysis of nasal sounds and nasalized vowels shows that the algorithm is able to yield a good spectral fit between the magnitude response of the branched tube model and the speech spectrum.