[mythtv-users] AudioOutputGraph progress

[John Pilkington]

On 12/02/2021 09:45, Mark Kendall wrote:
> On Thu, 11 Feb 2021 at 16:54, John Pilkington <johnpilk222 at gmail.com> wrote:
>>
>> Hi Mark :  Thanks for looking at this (among much else).  I have b15f981
>> running in F32 and have looked at a recorded GTS (Greenwich Time Signal)
>> on BBC Radio 4.  A good AOG could be helpful for editing radio, while
>> for video the picture provides the most important editing clues.
>>
>> In radio the positioning isn't accurate or repeatable - and the display
>> timescale varies too.  My recording began 2 minutes early, and the
>> 'imaginary' video frame rate is 25 fps.  Hence the start of the final
>> pip should be at 3000 frames, and the start of the first at 2875.
>>
>> Stopping playback at the first audible pip shows a frame count around
>> 2810, while the first to show in the AOG has been around 2720 and the
>> last around 2830.  In the past, only step sizes in radio down to 1
>> second have worked reasonably well.
>>
>> I tried making improvements in this some time ago, but my efforts seemed
>> more likely to break things than yield improvements.  I hope and expect
>> that you can do better :-)
> 
> Hi John,
> 
> The only functional changes I've made to the audiograph code to date
> were to fix the crash on macos. The subsequent changes were purely
> cosmetic.
> 
> In testing those changes, I pretty much decided that the cutlist
> editor is a bit of a mess at the moment. Using VAAPI for decoding,
> seeking was erratic in a number of ways and as such, if I were using
> the editor, I wouldn't be comfortable that the frame displayed is the
> correct one (note: software decoding is much better - though I
> wouldn't describe it as perfect). Likewise, although I can't be sure,
> it really doesn't look like the audio graph is showing consistently
> accurate data. I've never really used that functionality and so don't
> know whether it is a regression or not. I'm a little more familiar
> with the code now.

Yes, I'm using software decoding when in edit mode; it's the only option 
on this box (el7 3GHz core2duo), and using nvdec on the 
similar-but-with-nVidia F32 is more awkward.  The main inconvenience now 
is still the tendency to pause for 20 seconds when a return to playback 
is expected.  I suspect is that it was a latent bug in the AOG code. It 
didn't happen before October, and keystroke sequences can be adapted to 
make it less likely.   Thanks anyway...

  >
> Going forward, the player code is only part way through a sizeable
> re-factoring - and one of the aims is to significantly improve
> handling of audio only streams. So I'm hoping that much of the audio
> only/radio stream issues will start to resolve themselves. Likewise,
> once the current round of changes is done, I'll start looking at the
> decoder in a little more detail - which hopefully will start to
> resolve some of the minor seek issues.
> 
> All told - what I'm trying to say is - stick with it, there is a plan
> and I'm hoping it will all come together before 0.32 is released.
> 
> Thanks and regards
> Mark