[mythtv] "Extra Audio Buffering" setting

[Stephen Worthington]

On Fri, 10 Nov 2017 11:22:03 +1100, you wrote:

>
>On 10/11/17 08:26, John P Poet wrote:
>> On Thu, Nov 9, 2017 at 2:20 PM Peter Bennett <pb.mythtv at gmail.com 
>> <mailto:pb.mythtv at gmail.com>> wrote:
>>
>>     Hi Devs
>>
>>     There is a Frontend General Playback setting called "Extra audio
>>     buffering" which mentions certain types of tuners that cause crackly
>>     audio. Actually it enables setting DecodeExtraAudio which then maybe
>>     sets a flag called lowbuffers in the demuxer (class AvFormatDecoder).
>>     There is lots of complicated logic around whether or not that flag is
>>     set and in some cases its value is saved and reset.
>>
>>     The lowbuffers flag is only tested once in the system and if set the
>>     program buffers 100ms of video packets in
>>     AvFormatDecoder::GetFrame, so
>>     that audio is not left behind.
>>
>>     Bug #13172 describes how playback becomes extremely jerky if you
>>     adjust
>>     audio sync negatively more than a small amount. The solution is very
>>     simple, just buffer more than 100ms in AvFormatDecoder::GetFrame. I
>>     tested with 1000ms and then you can successfully set audio sync to
>>     somewhere around -1200 before it becomes jerky. In the patch for
>>     the fix
>>     I settled on a figure of 250 as a good default and added for a setting
>>     called AudioReadAhead for the value.
>>
>>     I would also like to get rid of the complex logic and flag surrounding
>>     the DecodeExtraAudio flag and replace that setting in frontend setup
>>     with AudioReadAhead with a default value of 250. With a value of
>>     250 or
>>     1000 it works on everything I have tried. There is no adverse impact
>>     that I can see, although it will take some extra memory for the
>>     additional buffering. I do not know why there is all the extra logic
>>     around whether or not to set the flag, it is safe and easy to set
>>     it all
>>     the time.
>>
>>     I propose a setting that specifies number of milliseconds of audio
>>     read-ahead with an explanation that says to increase this value if the
>>     audio is jerky especially when adjusting audio sync negative.
>>
>>     Any comments?
>>
>>     Peter
>>
>>
>> I have not looked at any of this code, but my guess is that it was to 
>> handle frame-grabber cards, where Myth had to mux the audio and video 
>> together itself.  We pretty much don't support those anymore (using 
>> one in this age is just asking for pain).
>>
>> If you want to play it safe, create a branch and ask as many people as 
>> possible to try it.  I will.
>>
>> However, if it is working well for you, even with your RPi units, then 
>> I expect it to be fine.
>>
>> John
>>
>Beware of and test with mythmusic and audio only streams when you change 
>this.
>
>Note that audio is rendered as it is seen in the decoded stream whereas 
>video is not.
>Audio only for mythmusic has as special shortcut that limits the buffer 
>to 500ms or less so that when you seek or change streams the response is 
>quick.
>
>You are correct in that the audio buffer size directly relates to the 
>amount of -ve sync achievable. This is a design constraint.
>
>Also some OTA streams have largely varying A to V time gaps on PTS and 
>the buffer is also required to absorb this. mkv's can also be quite 
>un-aligned Ive found.
>
>If you enable some of the more fancy features in audio land such as 
>ch->6ch upsampling and AC3 reencoding, you will also chew up the buffer. 
>The upsampling is especially a buffer hog partially due to the LPF for 
>the subwoofer channel. I have noticed this is quite long when I 
>start/skip for audio to catch up (600+ms).
>
>Changing the extra buffering to a number in settings is ok but you need 
>to add some help text.
>That said, I have patches to change mine to 1.5 secs as default audio 
>buffer size because the existing is inadequate for some of my media.
>
>As an aside, there is also a bug in passthru mode in that if the rate 
>does not match the standard AC3 rate, then you get no sound out of your 
>digital amp as the AC3/DTS stream is not framed correctly due to 
>differing bitrates on the SPIDF line. The workaround is to enable 
>timestretch as this reencodes correctly and then all sound is good. This 
>happens on most of our (AU) HD channels.
>
>HTH
>mark

When playing downloaded video files, be aware that there is a nasty
bug in one of the most commonly used programs used to make these
files, Adobe Premiere.  Whenever it creates slow motion in its files,
the audio/video interleaving seems to be disabled, at least for the
slow motion section.  So you get all the video frames followed by all
the audio frames, with no interleaving.  If the slow motion section is
small enough, the playback program's buffering can cope with this, but
MythTV seems to have a smaller buffer size than most Windows player
programs, and that means it will fail to play back these files when
the Windows software is OK.  This causes MythTV to be blamed for the
problem, rather than Adobe.  I have tried to convince the owner of one
web site that I subscribe to that it is an Adobe bug he should report
to them and get fixed, but as he did not have sufficient technical
knowledge I was unable to convince him and he just told me to use a
better program to play the files.  My solution is to run his files
through ffmpeg with the right options to get them re-interleaved
correctly.  But it would be good to work out what sort of buffer sizes
are commonly used in other software and make MythTV use at least that
much buffer, so it does not get blamed for problems like this.