Create Ava: an intelligent, full-body, controllable human avatar that can hear/see and is capable of natural, spoken conversations with humans.
Issues Involved or Addressed
Current virtual personal assistants like Alexa, Siri, Google Assistant, Cortana most commonly follow a “voice-in-a-box” approach: unimodal voice I/O. In addition, interactions are limited to narrow domain, single turn dialogues: “what is the weather forecast for today?”, “play Easy on Me”. The Ava digital human will be an intelligent multimodal voice+vision full body digital human avatar capable of multi-turn open-domain interactions. The AVA Lab is creating the Conversational AI technologies for AVA including speech recognition/synthesis, natural language understanding/generation, dialogue management. To complement the AVA Lab, this VIP will address the challenge of creating a full-body, fully controllable high fidelity human avatar to represent Ava. This work of this VIP will involve advancing the state-of-the-art in deep learning methods to train Text-to-Face and Text-to-Body animation models.
Methods and Technologies
Academic Majors of Interest
Preferred Interests and Preparation
software development skills (familiarity with Python), strong aptitude to learn advanced experimental methods, applied mathematics, statistical modeling, machine learning
Meeting Schedule & Location
- Machine Learning