Constructing a Japanese Multimodal Corpus From Emotional Monologues and Dialogues

摘要

To fully incorporate emotion into human-computer interaction, rich sets of labeled emotional data is prerequisite. However, in Japanese, the majority of the existing emotion database is still limited to unimodal and bimodal corpora. To allow more complete observation of emotion occurrences, we construct the first audio-visual emotion corpora in Japanese, collected from 14 native speakers. Preliminary speech emotion recognition experiments on the corpus and achieved an accuracy of 61.42% for five classes of emotion.