-
Notifications
You must be signed in to change notification settings - Fork 0
/
README.playout
61 lines (52 loc) · 2.91 KB
/
README.playout
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
>>>Quick question on rat: does your adaptive playout buffer algorithm take
>>>the redundancy into account? I would expect it would do something like:
>>>compute the playout delay as normal, and then add N*packetization_delay,
>>>where N is the number of packets in the past covered by the redundancy
>>>in a packet. Is that basically right?
>>
>>To avoid modifying the playout calculation for each channel coder the
>>channel coders may add the extra delay in their decode stage. We have
>>two buffers in the receiver: one for media data and one for channel
>>data. The same playout time is used for both. When a channel coded
>>unit reaches playout it is passed to the channel decoder. The
>>channel decoder may or may not output a media unit. If it does it is
>>decoded and played. In the case of the redundant decoder there is
>>some internal buffering inside (N * packetization_delay) to ensure the
>>redundant data can be used.
>
>What is a channel coder in this context? Are you talking abut the
>redundant codecs?
There is a generic framework for channel coders. We currently have
redundancy and layered, and used to have interleaving, though it was
removed for some re-engineering and I have not had time to add it
back. It should work with parity and Reed-Solomon also.
>Wouldn't the additional buffering need to be in the playout buffer of
>the main codec?
Right, there are two playout buffers, one for the channel coded data
and one for the media data. The channel coder can output media units
with any playout offset it chooses. The story in the case of
redundancy goes as follows: let the channel unit (redundant packet)
playout time be Tc, then media units are output at:
Tm (i) = Tc + max_offset + i * frame_duration
where Tm(i) is the output time of media frame, i, arising from the
channel unit.
>My question is also how this amount and the playout
>delay computed from network jitter are used - do you add the
>n*packetization delay to the playout delay computed from jitter?
The playout delay calculation has some caveats, but it is basically 3
* jitter as calculated at the arrival time of the first packet in the
talkspurt.
Caveats are:
1) with first packet received we have no jitter information so take
minimum of 3 * frame_duration and 3 * 20ms.
2) if user has set bounds on the minimum and maximum playout delay,
constrain playout to sit inside these bounds.
3) if lip synchronization is in use, use playout negotiated with
video tool (always >= jitter based playout calculation).
4) playout time of incoming audio overlaps with audio currently
being played - push back.
5) playout delay is less than cushion, use cushion - cushion is
estimate of how much audio is in device buffer to protect
against process losing cycles in scheduler.
>>There is no option to vary the amount of internal buffering for
>>redundancy, but obviously it would be trivial to do.