|
From: Stefan L. <Ste...@sn...> - 2001-11-17 23:15:41
|
On Saturday 17 November 2001 22:56, Erik Walthinsen wrote: > On Sat, 17 Nov 2001, Stefan Lucke wrote: > > So does that mean we may not trust the given value of samples per frame ? > > Do we need a on the fly resamlper to adjust such differences for playing > > ? > > IIRC, the IEC spec states that the number of samples per frame can vary > within certain bounds, especially for certain types of audio. I don't > have the spec (Buck does), so I'm not sure how exactly it deals with these > mismatches. They are caused by unmatched hardware clocks in the recording > device, and are common just about everywhere. One of the problems in PC > hardware is that the vertical retrace and the audio clocks are almost > never truly in sync... ;-( Libdv doesn't use hardcoded numbers of samples. From audio.c numbers may vary between 1264 and 1296 samples per frame for PAL at 32000Hz (31600 to 32400). But those from dv header calculated values are perhaps wrong. Did you solve such fine tuning issues with GStreamer ? I know that there are some issues har to resolve. Recording at 31993 instead of 32000 and playback at 32007 implies that you are off by 1 second within 2/3 of an hour. For grabdv and kino it is very hard since they gernerate DV2 type avi files on the fly. It seems to be impossible to do that without audio decoding and filering such invalid samples. -- mfg Stefan Lucke (Ste...@ep...) |