A fractal approach to normal and pathological voices.

Using the box-counting method, we demonstrated recently that the stationary signal of vowels is not fractal, but provides the opportunity to design in the smallest scale a kind of signature for each vowel. This fractal approach to these components of speech allows us to quantify the roughness of the voice, between I (sinusoidal complex signal) and 2 (white noise). We used this method to compare these values in normal and pathological voices. We studied the speech of 10 normal speakers, 6 patients suffering from unilateral vocal fold palsy and 6 suffering from various other dysphonias. The meaning of this fractal measurement is discussed and compared with electroglottogram and spectrographic analysis.