Which emotional states in SER rely on much finer variations in timing and spectral tilt compared to anger?

Boredom or confusion (Lower-arousal states)

The progress in Speech Emotion Recognition (SER) reveals an imbalance in classification difficulty based on the arousal level of the emotion being targeted. High-arousal states, such as anger or excitement, generate easily detectable and pronounced acoustic shifts—their loudness increases, and their pitch often elevates significantly, making them relatively easier for algorithms to map reliably. Conversely, lower-arousal or more subtle affective states, specifically cited examples being boredom or confusion, manifest through much finer, less obvious variations in vocal characteristics. These subtle states depend heavily on minute changes in timing characteristics (like slight hesitations or pauses) and spectral tilt (the distribution of energy across frequencies). Therefore, achieving high accuracy in recognizing these quiet, less acoustically marked emotions often requires significant innovation in loss functions or feature extraction layers to achieve the necessary sensitivity.