Real-Time Human-Music Emotional Interaction Based on Deep Learning and Multimodal Sentiment Analysis

Figure 6

Schematic diagram of the process of feature-level fusion (a) and decision-level fusion (b) methods used in multimodal fusion. The feature level fusion concatenates the feature vectors of all modes and then inputs them into the classifier, whereas the decision level fusion inputs the feature vectors of different modes directly into the classifier and aggregates the feature scores according to different weights.

(a)

(b)