In our ACL 2026 paper, we introduce AV-Dialog, the first full-duplex spoken dialogue model that combines audio and visual input to converse in noisy, multi-speaker environments.
Paper: aclanthology.org/2026.acl-lo…
Webpage: avdialog.cs.washington.edu/
2. Explicit turn-taking supervision is crucial. A dedicated control stream that predicts when to speak before determining what to say enhances response-timing accuracy in noise.
1
24
Today’s AI assistants passively wait for questions. But what if they could anticipate when to help-without explicit user invocation? Meet LlamaPIE, the first real-time proactive assistant to enhance conversations via discreet, concise guidance delivered by hearable.#acl2025
We bring LLAMAPIE to life in a fully integrated speech-to-speech setup, worn by users via a bone-conduction headset. We designed mock interviews and trivia tasks, where users engaged in casual conversations with several topic-specific questions.
1
88
