Papers by Yinglin Zheng
Predicting Turn-Taking and Backchannel in Human-Machine Conversations Using Linguistic, Acoustic, and Visual Signals (2025.acl-long)
Copied to clipboard
| Challenge: | Existing systems for human-machine conversations are limited in predicting turn-taking and backchannel actions. |
| Approach: | They propose a multi-modal face-to-face (MM-F2F) human conversation dataset . they collect and annotate over 210 hours of human conversation videos . |
| Outcome: | The proposed model achieves state-of-the-art on turn-taking and backchannel prediction tasks. |