[FFmpeg-user] Create an AAC stream matching the Core Media Audio packet format / priming etc?

[Marton Balint]

On Mon, 19 Jun 2017, Mark Burton wrote:

>> On 15 Apr 2017, at 13:25, Christian Ebert <blacktrash at gmx.net> wrote:
>> * Marton Balint on Saturday, April 15, 2017 at 12:25:58 +0200
>>> On Sat, 15 Apr 2017, Christian Ebert wrote:
>>>> * Marton Balint on Saturday, April 15, 2017 at 07:55:22 +0200
>>>>> Last time I checked (a year ago or so), ffmpeg created a correct .mov
>>>>> edit list to signal the audio priming.
>>>>> 
>>>>> https://developer.apple.com/library/mac/documentation/QuickTime/QTFF/QTFFAppenG/QTFFAppenG.html
>>> 
>>> Hmm, a recent fix changed one of the hunk contexts... Could you try
>>> this new attached patch?
>> 
>> I'm afraid for my purpose - segmenting - it does not make a
>> difference. I still get the same overlong first segments;
>> depending on codec - native aac has much shorter 'delays' than
>> fdk-aac, they are minute, but they still make the hls muxer set a
>> 1 second higher TARGETDURATION, even when all segments are of the
>> exact duration up to the sixth decimal - this can be corrected
>> by hand because the spec tolerates such minimal divergences.
>> 
>> But again, for the purpose of segmenting it has not any effect.
>> 
>> It may well fix Mark's use case. I haven't checked that.
>
> Hi Marton,
>
> I wasn’t able to test your patch myself, but another very helpful 
> developer did apply it and ran a test for me. The result was much better 
> sync with your patch, although still not a complete match for the input 
> file’s sync when decoded in Quicktime.
>
> Adding the 'roll' sgpd atom and sample-to-group atoms does appear to be 
> a requirement for AAC priming to be better understood by Quicktime. 
> Although there is still something else happening to scupper perfect 
> sync, it would be useful to apply this patch if possible.
>

Can you measure how many samples are the delay before and after my patch? 
I'd rather do some additional testing now, instead committing something 
that is only half-correct.

Thanks,
Marton