Figure 2.
A diagram presents hierarchical emotion distribution representations and an extraction process. On the left, hierarchical emotion distributions are shown at three levels, including utterance level, word level, and phoneme level, aligned with a phoneme sequence and corresponding raw text reading I love you. Each level contains multiple emotion distribution units, and the hierarchical distribution is described as having a shape of 12 by 6, with each individual emotion distribution having a shape of 4 by 1. Example emotion intensity values are listed as angry 0.12, happy 0.50, sad 0.68, and surprise 0.95. On the right, a hierarchical emotion distribution extractor shows phoneme level, word level, and whole utterance audio processed through segmentation, feature extraction, and ranking functions. These ranked emotion distributions are combined to produce a final hierarchical emotion distribution, which outputs the same set of emotion intensity values for angry, happy, sad, and surprise.(a) Example of Hierarchical Emotion Distribution (ED) including EDs at levels of utterance, words, and phonemes; (b) Diagram of Hierarchical ED Extractor