This project introduces EMo-Mask, an extension of the motion generation model MoMask, which incorporates emotion understanding to generate diverse and expressive human motions. By integrating an EmotionEmbedder module, EMo-Mask learns meaningful representations of emotions and guides the motion generation process. Experiments demonstrate that incorporating emotion embeddings at the M-transformer input of MoMask yields the best performance. The project identifies limitations, such as the sensitivity of average Mean Squared Error loss, and proposes future directions for enhancing the EmotionEmbedder architecture and exploring perceptually relevant loss functions. EMo-Mask represents a significant step towards generating emotionally expressive motions, contributing to the creation of more engaging and believable animated characters.
