Detecting happy and sad exclamations in Mandarin with acoustic features

摘要

Emotion recognition has been a hot and significant topic in Natural Language Processing, human-computer interaction as well as sentiment analysis technology. Exclamation, either happiness or sadness, is a very typical sentence type to express strong emotions in everyday communication, social medias, customers' responses to services and products and cross-cultural communication. Moreover, both types of exclamation can share the identical literal form but express distinctively different emotions by means of linguistic contexts and intonation devices. However, not much is known about their subtle but unique speech features. To shed light on this demanding issue for sentiment analysis and human-computer reaction, the present study explores the two facets of exclamation in Mandarin Chinese (Putonghua) in order to probe into the acoustic features. The findings reveal that happy exclamation differs from sad exclamation in their overall sentence duration, subpart intonation phrase duration, average sentence pitch level, maximum pitch as well as pitch range. These prime distinctions in duration and pitch parameters provide insights into robust AI -based emotion recognition of speech.