Real Time Multimodal Conversational AI

The main objective of the workshop is to bring together diverse communities to discuss real-time multimodal conversational AI. This means that even negative results, if relevant, will be considered. Furthermore, submissions focusing on black box systems or minor hyperparameter tuning for incremental improvements to some metrics are not encouraged. However, papers presenting incremental technical novelty are allowed. Finally, papers with a multimodal bent will be preferred. For instance, this includes speech recognition for conversational AI, expressive text-to-speech or spoken language understanding.

The workshop welcomes the submission of work on, but not limited to, the following research directions: 

• multimodal representation learning for conversational AI 

• full-duplex multimodal conversational models 

• social intelligence and communication for digital and embodied agents 

• egocentric multimodal perception 

• multi-speaker conversation 

• real-time multimodal processing for interactive systems 

• human-machine interaction in physical environments 

• predictive turn taking 

• speech/text controlled expressive gesture and facial expression synthesis  

• datasets and benchmarks relevant to multimodal conversational AI.