Speech emotion recognition has become crucial in the field of human computer interaction for important applications like early detection of Alzheimer's Disease- (hereafter AD). The aim of the present paper is to introduce a new feature for emotion detection from speech using the Multi Fractal Detrended Fluctuation Analysis (MFDFA) method. We propose a quantitative parameter for categorizing various emotions by analyzing the non-stationary details of the dynamics of speech signal, generated out of differing emotions. The observed contrast in this parameter on Angry and Sad emotion can be used for diagnosis of AD. It will be a positive step towards early detection of AD using speech analysis.
Susmita Bhaduri, Rajdeep Das and Dipak Ghosh