Non-linear techniques for dysphonic voice analysis and correction

This paper aims at finding suitable parameters for dysphonic voice analysis and classification. Moreover, a non-linear noise reduction scheme is proposed, for voice correction. Typical quantities from chaos theory and some conventional ones are evaluated, in order to provide entries for feature vectors in a feature space. Geometric signal separation is applied for voice classification, by means of a properly defined 'healthy index'. This allows moving the feature vector of pathological voices from the sick region to the healthy one, in order to enhance voice quality. The practical advantage is twofold: first, physicians are provided with a better understanding of voice dysfunctions for surgical and rehabilitation purposes, second non-invasive devices for voice denoising could be build up as an aid for dysphonic people.