Virtual Human Toolkit

The ICT Virtual Human Toolkit is a collection of modules, tools, and libraries designed to aid and support researchers and developers with the creation of virtual human conversational characters. The Toolkit is an on-going, ever-changing, innovative system fueled by basic research performed at the University of Southern California (USC) Institute for Creative Technologies (ICT) and its partners.

Designed for easy mixing and matching with a research project’s proprietary or 3rd-party software, the Toolkit provides a widely accepted platform on which new technologies can be built. It is our hope that, together as a research community, we can further develop and explore virtual human research and technologies.

The ICT Virtual Human Toolkit is built upon a common modular architecture which enables users to utilize all modules as is, one or more modules coupled with proprietary components, or one or more modules in other existing systems. Our technology emphasizes natural language interaction, nonverbal behavior and perception. Its main modules are listed below. See Documentation for an overview of the architecture, the messaging API, and other components.

MultiSense: A multimodal sensing framework which is created as a platform to integrate and fuse sensor technologies and develop probabilistic models for human behavior recognition. MultiSense tracks and analyzes users’ facial expressions, body posture, acoustic features, linguistic patterns and higher-level behavior descriptors (e.g. attention, fidgeting). It uses the Perception Markup Language (PML). NPCEditor: At the core of the NPCEditor is a statistical text classification algorithm that selects the character’s responses based on the user’s utterances. A character designer specifies a set of responses and a set of sample utterances that should produce each response through a provided authoring tool. The NPCEditor also contains a dialogue manager that specifies how to use the classifier results.