Emotion Recognition on Real Life Emotions - Robust Emotion Recognition Using Spectral and Prosodic Features - page 96

Fig. 6.1 Training and testing phases of developing emotion recognition systems
is given in Chap. 2 . The performance of ERSs on IITKGP-MESC is shown in the
last column of Table 6.3 . The combination of LPCC and formant features has per-
formed quite well for the real life-like emotion recognition task.Real life emotion
recognition performance is highest in case of pitch synchronously extracted spectral
features. The trend of results is almost the same as in the case of the other two data-
bases. Analyzing the results of three given speech corpora, the emotion recognition
performance is considerably high in case of real-life emotions, mostly because the
emotions are expressed with proper discrimination, as the movie audio clips contain
sufficient linguistic and contextual information.
Real-life emotions are further investigated using other speech features. Emotion
recognition studies are conducted on IITKGP-MESC using source, system, and
prosodic features. Table 6.4 shows the emotion recognition performance using
source, system, and prosodic features individually and in combination. From the
results presented in the Table 6.4 , it may be observed that there is a considerable
improvement in the emotion recognition performance for real life emotions using
spectral features, compared to their source and prosodic counterparts. However, a