LPC Spectrum First Peak Analysis for Voice Pathology Detection

AbstractThe detection of voice pathologies using speech processing techniques can be a useful contribution for the diagnose of larynx diseases. The main objective of this work was to inspect the spectral envelope of the voice signal searching for information that allows voice pathology detection. The frequency and bandwidth of the first peak from the spectral envelope obtained from Linear Predictive Coefficients (LPC) of 30th order was found to be a valuable feature being used for voice pathology detection. In the corpus considered in this work we obtained a 100% discrimination between healthy and unhealthy subjects and a 87% discrimination between nodules and Reinke's edema