has anyone tried using video to help with end of t...
# agents
i
has anyone tried using video to help with end of turn detection? My environment is very noisy, so I’m thinking that using lip movement detection + VAD to determine eot would be better, curious if anyone has explored this?
r
I've read some research papers on it in the past but I've not tried it. This maybe helpful if you need face landmark detection: https://github.com/livekit/python-sdks/tree/main/examples/face_landmark
i
do you happen to remember which papers you read? Would love to check them out
r
I don’t recall off hand. There was a burst of papers on the topic earlier this year. I just did a quick search for an example paper and this is one example https://arxiv.org/html/2505.21043v2 I think it is an interesting thing to look into but I was more interested in other areas of turn detection so did not pay to close attention to them.