Replies: 2 comments
-
couple of ideas... first, make sure to squeeze every bit of performance - run detection in a webworker and make sure that any unnecessary modules are disabled (iris, description, emotion, etc.) second, dont treat each detection separately third, i can take a look at the actual implementation for accuracy - been a while since i've implemented this. |
Beta Was this translation helpful? Give feedback.
-
Thank you Vlad. Good tips. |
Beta Was this translation helpful? Give feedback.
-
Hi Vlad,
Do you have any recommendations on how to get better results out of the "Mouth % Open" gesture? I am trying to determine if someone is speaking and use the microphone volume paired with the percentage of mouth open (can't rely just on microphone volume as it could be system audio, etc). It works reasonably well, but there are quite a lot of times when there's no % returned (or the opposite, it thinks the mouth is open when it's really not). Just wondering if you had any thoughts on ideas to explore for this? Other than modifying your 10% threshold, anything else you can think of? Thanks!
Beta Was this translation helpful? Give feedback.
All reactions