ActorMind Framework for Speech Role-Playing Released
Researchers have introduced ActorMind, a multi-agent reasoning framework for speech role-playing, along with ActorMindBench, a benchmark for evaluating such systems. The work addresses the limitation of current role-playing research to text-only modalities by incorporating speech, which is central to human interaction. ActorMindBench is a hierarchical benchmark comprising 7,653 utterances across 313 scenes and 6 roles. ActorMind is an off-the-shelf, chain-of-thought style framework that emulates human actor reasoning to generate spontaneous responses with personalized verbal traits based on role, scene, and dialogue. The paper is available on arXiv under ID 2604.11103.
Key facts
- ActorMind is a multi-agent reasoning framework for speech role-playing.
- ActorMindBench is a hierarchical benchmark with 7,653 utterances, 313 scenes, and 6 roles.
- The framework emulates human actor reasoning using chain-of-thought style.
- It generates spontaneous responses with personalized verbal traits.
- The research addresses the gap of text-only role-playing by incorporating speech.
- The paper is published on arXiv with ID 2604.11103.
- ActorMind is described as off-the-shelf.
- The benchmark includes Utterance-Level, Scene-Level, and Role-Level content.
Entities
Institutions
- arXiv