Automatic Formant Tracking Method by Fourier and Multiresolution Analysis

This paper proposes two approaches for formant tracking using Fourier and wavelet ridges. This work aims to improve the performance of formant tracking algorithm. The speech signal is decomposed into time-frequency representations issued from windowed Fourier transform and wavelet transform. Formant tracking is achieved by exploring ridges from time-frequency representation. In these two methods we have introduced a continuity constraint based on the computation of center of gravity for a set of frequency formant candidates which leads to connect a frame of speech to its neighbours and thus to improve the robustness of track. These approaches are validated by their application on synthesized signals. The Fourier tracking method shows a reliable estimation in comparison with the values given in synthesis contrary to wavelet tracking method. In the same way, the two analyses are applied on natural voiced speech issued from an Arabic database formant labeled. The formant trajectories obtained by the two proposed algorithms are compared to those of the hand edited formant database as a reference. The manually Formant labeling is carried out used the Winsnoori tool. The comparison of the results to the two proposed approaches shows the adequacy of the Fourier ridges method.