Affective Agent Model

In the affective agent model, the agent is embedded in an environment and is interacting with an interlocutor. The agent receives inputs from both, which must be appraised. It then decides on an appropriate response and uses Affective Speech Synthesis (ASS) to generate a response.

From @triantafyllopoulosOverviewAffectiveSpeech2023

Pasted image 20240420185849.png

For the appraisal, the relevant parts of the input must be realized. This could be done by novelty, intrinsic pleasantness/valence, goal relevance, urgency and power/control.

Recent works suggest models that learn the response using learnt behaviors, like [[Reinforcement Learning (RL)]] and have been shown to increase the subjective score of richness.