Automatic Detection of Pathological Voices Using GMM-MLLR Approach

Modern lifestyles have increased the risk of suffering some kind of voice disorders. It is estimated that nearly 19% of the population have suffered from dysphonic voicing. It is very important to detect pathological voices automatically. Many classification methods have been used to detect the pathological voices automatically and got good results. In this paper, we focus on the automatic detection of pathological voices using GMM-MLLR approach. MLLR Transformation matrix of GMM model is shown to be an efficient feature of detecting pathological voices in our experiments. In the evaluation task, the EER of our test database composed by 141 pathological and 17 normal utterance is 8.2%.