Recognition of Speech Signals using Recurrence Nonlinear Dynamics ☆

In this analysis we investigate the recognition of speech signals in various extreme and correlated noisy conditions. We consider two different cases, (1) the speech signals of same person in a different noisy environments, (2) same speech of different people under noisy environment. It is not possible to recognize the aforesaid speech signals by the conventional frequency and time-frequency methods since the noise is highly correlated with the signals. We propose a method to recognize and distinguish their patterns in both the cases by using nonlinear recurrence dynamics. The recognition is based on the distribution of the diagonal lines which are parallel to the line of identity in their recurrence dynamics. Experimental results support our proposed analysis.