음성 감정인식을 위한 분류기에 대한 조사 및 비교분석

In this paper we describe classifiers used in speech emotion recognition, and summarize important experimental results provided by previous researchers. The results show that there is little difference in performance between classifiers with respect to the same features. On the other hand, the performance of a single classifier is improved by using a hierarchical structure. In addition when the speaker is known and a classifier is trained with the speaker’s training data, the recognition rate is increased to twice of the speaker-independent case. Based on these results, we propose a way to improve speaker-independent recognition.