Robert Kulagowski (rkulagowski@xxxxxxxxxxxxx): > I've tried both mp1e 1.9.2 and ffmpeg (CVS). Neither appears to > support alsa directly - both use the OSS emulation. (I think that I > read somewhere that Alsa was much better as far as providing a stable > audio signal for maintaining A/V sync. Don't know whether this is > true or not.) It's important to have timestamps on all audio fragments you get from the soundcard in something objective that you can relate to frames from the video input. Unfortunately, this is a challenge. There is no standard for timestamps from the kernel, so relating timestamps from v4l and ALSA is a pain as far as I know. In my recorder, I run as root at realtime priority and try to ensure a minimum amount of time between the kernel giving me the data and my thread receiving it, but I'm not sure what's best. > Is anyone aware of a capture program that can maintain sync over a > long period of time? I'm thinking that at a minimum 2 hours is > required. In my capture app, I use the timestamps I get after the full capture and do a linear approximation of the difference between the amount of audio received and amount of video. I can then use that to resample the audio when writing to the destination file. But my app isn't done yet. You could see about porting the code somewhere, since I haven't had much time, or at least steal the idea. The linear fit code should be in CVS. My code is in CVS here: http://www.sf.net/projects/reetpvr/ -- Billy Biggs vektor@xxxxxxxxxxxx
