Research

Interactive & Proactive AI

We build AI that perceives the world through vision, audio, speech, and language, understands what people are doing, and steps in at the right moment without being asked. Our agents listen and speak at the same time, follow the context of long interactions, and act through tools to help people in real-world tasks. Our research covers multimodal and omni-modal large language models, full-duplex spoken interaction, proactive assistance, conversational memory, and the evaluation of human-AI interaction.

Generative Models

We study generative models that create realistic and controllable content across images, video, audio, speech, music, and text. Our focus is on making generation higher in quality, faster, more controllable, and more personal, and on new model architectures that push what generative models can do.

Speech, Audio & Music

We develop AI that hears and understands every kind of sound, from human speech to environmental sounds and music, and that can generate them. Our research covers speech recognition and synthesis, voice conversion, speech enhancement and separation, speech translation, audio understanding, and music generation.

Page updated

Google Sites

Report abuse