Optimization for parameters of a vocal tract model based on simulated annealing algorithm
Lincan Zhou, Boyu Si, Ning Li, Zaoming Huang · 2010
Vocal tract can be modeled by analyzing the speech signals. The accuracy of this method depends on specifying and optimizing the articulatory parameters properly. A kind of vocal tract model, which is described in a midsagittal plane, generally contains 8-12 articulatory parameters, which have close relations with the acoustic formants. Calculating the first four formants frequency according to these articulatory parameters is in order to make mean error between natural formant and model formant as small as possible. It is a combinatorial optimization problem. This problem can be solved by simulated annealing algorithm. Mermelstein's model is a midsagittal vocal tract outline, which can match the X-ray tracings. This paper chooses an improved Mermelstein's model, proposed a method for optimizing the articulatory parameters based on simulated annealing algorithm in MATLAB. The results of experimentation show this method is effective.