Formant Tracking using Fixed and Adaptive Sine Liftering

Several formant tracking algorithms using fixed and adaptive sine liftering method are tested. The scheme of these algorithms consists of a preprocessor done by the standard linear prediction method, a bandpass liftering with the well known Juang-Rabiner-Wilpon sine window and finally formants tracking by solving for the roots of the linear prediction processed polynomial. The parameter of the sine window is fixed a priori or obtained with a Bayesian approach to model comparison of Gull and Skilling. We have found that it is desirable to use a short fixed liftering window to reduce the formant tracking error rate. The use of an adaptive sine window confirms the above result and improves the model generality although it does not lead to the most powerful solution in the sense of the mean error performance.