An Improved Approach Of Emotion Recognition Combining Spectral And Prosodic Features With Reference To Assamese Language

In the process of classification of emotions, it is seen that similar emotions always lead to misclassification. Such misclassification need to be reduced by taking an extra measure on performing early classification of those most confusing emotions into some different sub-groups. By grouping acoustically over lapping emotions into separate categories in stage one and classifying individual emotions in stage two. In the current study, “speaking rate” is chosen as the decisive factor for the sub-grouping of acoustically similar emotions. Later both Spectral and prosodic features in combination is used for further classification [1]. In stage one; by using spectral and prosodic features each emotion is categorized in 3 wider groups: 1) active 2) normal, and 3) passive emotions. Such wider groups are made based completely on speaking rate. In stage two, individual emotion sub grouping performed within each wider group. Figure 1.1: Duration of “ ” for emotions a) disgust, b) neutral and c) sarcasm Since it is observed that Excitation source feature has no appreciable influence in speech emotion recognition so it is avoided. We know that speaking rate is only a measure in utterance of number of syllables per unit amount time. It is accepted as most important characteristics for each speaker Abstract: The Speaking rate feature of speech can be explored for discriminating robust emotions. In real life, it is found that certain emotions are used to be very active with high speaking rate while some are very passive with low speaking rate. Keeping this motivation, a Phase II emotion recognition system has been proposed where three broad groups (active, neutral and passive) are taken in Phase I and each broad group are further classified in Phase II. In each stage classification of emotions are done by exploring Spectral and prosodic features. The combination of both spectral and prosodic features found to be performed better.