URL:

files/proceedings/2005/DesRocSMC05.pdf

Abstract:

In this paper we discuss the applicability of the Mellin transform for vowels recognition, focusing on spectral envelope scale distribution. The hypothesis used is that same vowels produced same spectral envelope shapes (same curves with different compression factor), so an energy, time and scale normalization can be used to map same vowels to same distributions. So, using fast algorithms, we study the applicability of this idea to build a realtime or quasi-realtime system capable of making vowel discrimination in a relatively straightforward way.