The Corpus Preparation and Effective Feature Representation of Emotional Speech

In this paper, we presented the corpus collection procedure and proposed the effective feature representation. We collected the emotional speech from 50 male and 50 female speakers and the corresponding statistics of the corpus was also demonstrated. The emotional speech corpus was further processed manually for the feature extraction and classification experiments. After introducing the feature generation techniques, PCA and LDA, the effective features representation for emotional speech is discussed. A further correlation analysis between the extracted features and the classes and classification experiments would be conducted to enhance the classification performance in the future.