Considering accuracies on the single class/emotion, there is an evident trend of higher accuracies for emotions with negative valence, namely Sadness and Anger.
← all excerpts
The Emotion Probe: On the Universality of Cross-Linguistic and Cross-Gender Speech Emotion Recognition via Machine Learning.
2
—
—
The sentences
However, there is a definite trend for feature domains, recurrent throughout all classification tasks, actually making up around 88% of the full feature sets (averaged by each task).