An exploration of feature vectors based on non-verbal emotion recognition

Emotions are a way through which people express their internal states, and good emotion recognition involves correctly identifying and interpreting these emotional expressions. It helps us to better understand the emotions of others, thus helping them to get better emotional value. Voice emotion recognition, as an inseparable part of emotion recognition, aims to perform voice emotion recognition by voice parameters using either existing publicly available datasets or homemade datasets, where the voice itself contains both speech and non-verbal parts. In this paper, a correlated positive feature vector for non-verbal emotion recognition is presented. The IEMOCAP dataset, the group of feature vectors meaningful for non-verbal emotion recognition is identified by improving the support vector machine correlation algorithm, combined with the weighted average weight calculation. The full-dimensional feature vectors are examined by combining the metrics of recall and accuracy. Thus, the positive direction three-dimensional feature vectors are filtered. Thirteen-dimensional MFCC 86.6%, F0 90.8%, and entropy value 80.1% of the three-dimensional positive direction feature vectors. These works can help to give direction to the subsequent work on feature vector extraction for non-verbal emotion recognition.