Re: [MMUSIC] Remove msid-semantic

[Harald Alvestrand <[e-mail]>]

Den 22. april 2015 14:15, skrev Magnus Westerlund:
> Martin Thomson skrev den 2015-04-21 19:05:
>> On 21 April 2015 at 02:31, Magnus Westerlund
>> <[e-mail]> wrote:
>>> However, in most cases one can't really play out
>>> all the buffered media, unless one can play it at faster than real-time.
>>
>>
>> I never imagined buffering being any more than a few milliseconds or
>> tens of milliseconds.  That might just be a case of allowing the audio
>> jitter buffer to fill.
>>
> 
> Well, that is fine for audio codecs which can generate good quality
> audio from any frame. But, for a video codec with its dependency on
> intra frames, then not buffering, in case the signalling arrives after
> the media, then discarding just the first packet will result in quality
> degradation or extra delay.  The actual situation will be the same as a
> packet loss situation (without possibility to repair) and recovery form
> it, most likely through a FIR request.
> 
> To be clear this only occurs in the situation where the signalling for a
> new track arrives after the actual delivery.

There's some simplification possible ... if the codec uses some variant
of pyramid encoding, where not all frames are required in order to
preserve the reference chain, all frames that aren't depended on can be
thrown away until the decision to show or not show can be made. When the
decision to show is made, just the frames required to have a proper
reference for the next incoming frame need to be encoded.

Probably increases reasonable hold times from milliseconds to seconds,
but not to minutes.....