This is a very complex and multi-faceted problem which requires many layers to interoperate, however, there are a few ways this could be achieved.

One problem I have is that USB devices cables' get physically disconnected just enough that the camera's light is still on, but the data cannot be transmitted. Perhaps Windows or another app could periodically try to take one second recordings (but of course don't send the recordings anywhere.)

Moving off of the 2.4Ghz WiFi band could help; this will take a while as consumer devices are slowly upgraded to support 5Ghz.

- Nvidia has AI-video compression which can reduce video to 0.11kb/frame (~100x smaller than regular video.) Lower bandwidth could allow for more reliable transmission: https://www.youtube.com/watch?v=eFK7Iy8enqM . Additionally, they have super-resolution which allows for lower-quality video to be upscaled.

- Video/audio decoding could be assisted with a thin-hypervisor layer or virtualization layer which dedicates a core or two of system resources to video decoding, and could abstract away some of the drivers from the program.

- Krisp.ai has very good noise reduction, but does require a monthly subscription to go past two hours per week.