Summary: | Humans sense, perceive, and convey emotion differently from each other due to physical, psychological, environmental, cultural, and language differences. For example, as recognized and studied by psychologists more than a century, it is easier for someone of the same culture to judge and recognize emotion correctly compared to those from different culture. In this chapter, we attempt to study the speech emotion recognition problem by using two speech corpora from the Berlin dataset and the NAW datasets. We have investigated the universality as well as diversity of two different cultural speech datasets recorded by German and American speakers, respectively. Experiments were conducted for identifying three basic emotions, namely, angry, sad, and happy with neutral as emotionless state from these datasets. MFCC coefficients were used as feature sets in the experiments, and MLP was employed as classifiers to compare the performance of these datasets. In addition, real-time recorded speech from drivers was also tested to see the performance in a vehicular setting. Finally, speech emotion profiling approach was introduced to explore the universality and diversity of the speech emotion features. © Springer Science+Business Media, LLC 2012.
|