[alsa-devel] What is the required accuracy of audio sampling rate?
Hi,
I have a PCI-e audio card that does not give very accurate sampling rate (about -0.2% error, i.e., if I set the sampling rate to 8K, I got only 7984 samples per second on average). It caused various synchronization problems in applications like gstreamer.
Since there will be some difference from one clock domain to another, I am wondering if there is any defined tolerance on the sampling rate, and where the right place is to correct this kind of error. I would really appreciate if someone can point me to the related documents/links.
Thanks, Yiliang
I have a PCI-e audio card that does not give very accurate sampling rate (about -0.2% error, i.e., if I set the sampling rate to 8K, I got only 7984 samples per second on average). It caused various synchronization problems in applications like gstreamer.
Can you be more specific on your sync issues? Generally the audio clock is used as reference, and video adjusts to it.
Since there will be some difference from one clock domain to another, I am wondering if there is any defined tolerance on the sampling rate, and where the right place is to correct this kind of error. I would really appreciate if someone can point me to the related documents/links.
It really depends on the standard, and if you are slave or master. Each standard has its own requirements. For example an AES-EBU interface will require 10ppm for professional uses as a source, and less as a sink. HDAudio requires the nominal BCLK 24MHz frequency to be between 23.9976 and 24.0024 MHz. I don't think on an I2S link there are such requirements, the frequency will depend on how accurate the audio PLL is. Note that the actual frequency is only one of the parameters influencing audio quality, jitter issues are much worse. If your clock is slightly off but remains stable, you are usually good. If it varies you'll have real audible quality issues. - Pierre
On 16 February 2010 23:58, Yiliang Bao yiliangb@gmail.com wrote:
Hi,
I have a PCI-e audio card that does not give very accurate sampling rate (about -0.2% error, i.e., if I set the sampling rate to 8K, I got only 7984 samples per second on average). It caused various synchronization problems in applications like gstreamer.
Since there will be some difference from one clock domain to another, I am wondering if there is any defined tolerance on the sampling rate, and where the right place is to correct this kind of error. I would really appreciate if someone can point me to the related documents/links.
There is no defined tolerance on sampling rate. Audio apps should be able to deal with a certain amount of error. xine uses the system clock for its synchronization, and resamples the audio output if the audio hardware is running at a different rate. Other apps might use the audio clock and sync to that. In that case, it might get the problems you describe. Try playing your media in xine and see if the problem disappears.
Kind Regards
James
On Wed, Feb 17, 2010 at 7:52 AM, James Courtier-Dutton james.dutton@gmail.com wrote:
On 16 February 2010 23:58, Yiliang Bao yiliangb@gmail.com wrote:
Hi,
I have a PCI-e audio card that does not give very accurate sampling rate (about -0.2% error, i.e., if I set the sampling rate to 8K, I got only 7984 samples per second on average). It caused various synchronization problems in applications like gstreamer.
Since there will be some difference from one clock domain to another, I am wondering if there is any defined tolerance on the sampling rate, and where the right place is to correct this kind of error. I would really appreciate if someone can point me to the related documents/links.
There is no defined tolerance on sampling rate. Audio apps should be able to deal with a certain amount of error. xine uses the system clock for its synchronization, and resamples the audio output if the audio hardware is running at a different rate. Other apps might use the audio clock and sync to that. In that case, it might get the problems you describe. Try playing your media in xine and see if the problem disappears.
The -0.2% error is cumulative, not random. After a three hour movie your audio will be off by 20 seconds.
Best solution is the get the audio clock synched as closely as possible to the video one. Any remaining error has to be dealt with in software like xine does.
Kind Regards
James _______________________________________________ Alsa-devel mailing list Alsa-devel@alsa-project.org http://mailman.alsa-project.org/mailman/listinfo/alsa-devel
Hi,
Thanks for all the replies. To reply Pierre's question:
Can you be more specific on your sync issues? Generally the audio clock is
used as reference, and video adjusts to it.
I played the audio using gstreamer alsasrc and alsasink. The audio plays well at the beginning, and gradually the sound becomes broken, and eventually unlistenable. If I play the audio along with video, there is a similar issue. In addition to the problem in the audio channel, the video will become intermittent as well after some time. I think the gstreamer is dropping video frames trying to synchronize the video with audio.
Similar to what is suggested here, I did some experiment on the audio driver so that its output is synchronized to the kernel clock, and the problems are gone (at least for the testing period of 2 hours). This confirms that the issues were indeed caused by inaccuracy in audio sampling rate and it appears gstreamer is not adaptive enough to accommodate such sampling error.
Thanks, Yiliang
On Wed, Feb 17, 2010 at 5:03 AM, Jon Smirl jonsmirl@gmail.com wrote:
On Wed, Feb 17, 2010 at 7:52 AM, James Courtier-Dutton james.dutton@gmail.com wrote:
On 16 February 2010 23:58, Yiliang Bao yiliangb@gmail.com wrote:
Hi,
I have a PCI-e audio card that does not give very accurate sampling rate (about -0.2% error, i.e., if I set the sampling rate to 8K, I got only
7984
samples per second on average). It caused various synchronization
problems
in applications like gstreamer.
Since there will be some difference from one clock domain to another, I
am
wondering if there is any defined tolerance on the sampling rate, and
where
the right place is to correct this kind of error. I would really
appreciate
if someone can point me to the related documents/links.
There is no defined tolerance on sampling rate. Audio apps should be able to deal with a certain amount of error. xine uses the system clock for its synchronization, and resamples the audio output if the audio hardware is running at a different rate. Other apps might use the audio clock and sync to that. In that case, it might get the problems you describe. Try playing your media in xine and see if the problem disappears.
The -0.2% error is cumulative, not random. After a three hour movie your audio will be off by 20 seconds.
Best solution is the get the audio clock synched as closely as possible to the video one. Any remaining error has to be dealt with in software like xine does.
Kind Regards
James _______________________________________________ Alsa-devel mailing list Alsa-devel@alsa-project.org http://mailman.alsa-project.org/mailman/listinfo/alsa-devel
-- Jon Smirl jonsmirl@gmail.com
participants (4)
-
James Courtier-Dutton -
Jon Smirl -
pl bossart -
Yiliang Bao