4 ms·
Couldn’t this be done by using some ML on the video stream?
by frutiger 5y ago
Couldn’t this be done by using some ML on the video stream?
- knorker 5y agoProbably the user experience would not be as good, though, since it'd be an audio-only feature that only works if you've enabled video.
- David 5y agoThat's way more computationally expensive and only available if the camera is on (and the user's mouth is visible).
- frutiger 5y agoThat’s true. I’ve never tried it with the camera off.
- lm28469 5y ago- Is device X receiving input vs - Training models, wasting years of CPU cycle world wide, &c. For such a dumb feature... I just don't see the point