The paper shows the methods for imaging power spectral density of speech and extracting formant frequencies from continuous voice. The methods will be used to improve the patients’ rehabilitation after the total laryngectomy surgery. The adaptive algorithms and transversal filters were implemented to estimate the transfer function of human vocal tract model. The estimation methods were based on statistical, Auto-Regressive model of speech production. The pilot study on formant frequencies, especially F1 and F2 formants, and their linear separation for each vowel has been presented. The method for recognition pathological voice has been proposed.

Index Terms. Speech signal spectrum analysis, adaptive filters, formant tracking and total laryngectomy