Formants Measurement for Esophageal Speech Using Wavelets with Band and Resolution Adjustment

Patients who have undergone a laryngectomy as a result of larynx cancer have extremely low intelligibility. This is due to the removal of their vocal fold, which forces them to use the air flowing through the esophagus: this is known as esophageal speech. Measurement of formant, in vowels, in a speech signal presents significant errors when conventional techniques such as linear prediction coefficients (LPC) are used. In this paper, an accurate formant location algorithm is presented. Taking the wavelet transform as base technique, each formant subband is approximated. After locating the approximated subband, a triband analysis is applied over it. With the accurate calculation of formants, it will be possible to apply techniques which improve voice quality in correction: the position of those formants and the instabilities that present their temporal evolution in esophageal voices