학술논문
Emotion classification of spontaneous speech using spoken term detection
Document Type
Conference
Author
Source
2017 IEEE 6th Global Conference on Consumer Electronics (GCCE) Consumer Electronics (GCCE), 2017 IEEE 6th Global Conference on. :1-5 Oct, 2017
Subject
Language
Abstract
This paper proposes an emotion classification method for spoken utterances using a spoken-term detection (STD) method. This is a keyword extraction method using spoken utterances. The extracted keywords are used to decide on the emotion category of an utterance. Most keywords extracted by the STD system are redundant and some of them negatively affect the emotion classification performance. Therefore, we propose a keyword filtering method based on the semantic relationships between keywords. The semantic relationships are calculated based on a word-embedding technique. The emotion classification results show that our proposed method outperformed the classification performance of the baseline system in which STD and keyword filtering were not used.