|
From: Justin S. <ju...@ex...> - 2003-09-05 07:43:19
|
Heikki Lindholm wrote: > Hullo, > could anyone clarify how the audio dropping algo is supposed to work. > There's one thing I don't like: video is dropped only because there's > difference between cheap capture cards clock (which value can't be read) > and system clock used to take timestamps. I get frame drops every 60000 > frames or so and would definitely like to capture more without drops. My > solution was to add epoch to video capture core (capture, say, 1000 frames > and then virtually start over). That should detect real frame drops nicely > while not getting confused by system clock vs. video clock drift. The > audio dropping algo however seems to use the constant 'real world' frame > period as a sort of frame counter, so obviously audio capture would also > need to be modified, but I just wondered why doesn't the frame estimator > (fvtime calculation) ask number of frames from video capture core? The whole problem is Linux's multimedia support. nvrec was written when there was still the intention of adding a system timer to Linux. The source of the timer was to be configurable (i.e. system time, or multimedia I/O, etc.). So I designed around 1 authoratative timer, and force everything else to match that timer. In the ideal world, this timer would have been configured to be that of the video capture core. Linux never got the timer support, so I never went any further with these plans... As for epoch support for audio, I would need to know exactly how you did the video support... > btw. I also made an ALSA core and could put it somewhere if anyone's > interested. Very interested! If you could submit a patch against the latest CVS of nvrec, I would appreciate it! -justin |