Clocking
When playing complex media, each sound and video sample must be played in a
specific order at a specific time. For this purpose, GStreamer provides a
synchronization mechanism.
Types of time
There are two kinds of time in GStreamer. Clock time is an absolute time. By contrast,
element time is the relative time,
usually to the start of the current media stream. The element time
represents the time that should have a media sample that is being
processed by the element at this time. The element time is calculated by
adding an offset to the clock time.
Clocks
GStreamer can use different clocks. Though the system time can be used
as a clock, soundcards and other devices provides a better time source. For
this reason some elements provide a clock. The method
get_clock is implemented in elements that provide
one.
As clocks return an absolute measure of time, they are not usually used
directly. Instead, a reference to a clock is stored in any element that needs
it, and it is used internally by GStreamer to calculate the element time.
Flow of data between elements and time
Now we will see how time information travels the pipeline in different states.
The pipeline starts playing.
The source element typically knows the time of each sample.
Sometimes it
is a parser element the one that knows the time, for instance if a pipeline
contains a filesrc element connected to a MPEG decoder element, the former
is the one that knows the time of each sample, because the knowledge of
when to play each sample is embedded in the MPEG format. In this case this
element will be regarded as the source element for this discussion.
First, the source element sends a newsegment event. This event carries information
about the current relative time of the next sample. This relative time is
arbitrary, but it must be consistent with the timestamp that will be
placed in buffers. It is expected to be the relative time to the start
of the media stream, or whatever makes sense in the case of each media.
When receiving it, the other elements adjust their offset of the element time so that this
time matches the time written in the event.
Then the source element sends media samples in buffers. This element places a
timestamp in each buffer saying when the sample should be played. When the
buffer reaches the sink pad of the last element, this element compares the
current element time with the timestamp of the buffer. If the timestamp is
higher or equal it plays the buffer, otherwise it waits until the time to
place the buffer arrives with gst_element_wait().
If the stream is seeked, the next samples sent will have a timestamp that
is not adjusted with the element time. Therefore, the source element must
send a newsegment event.
Obligations of each element.
Let us clarify the contract between GStreamer and each element in the
pipeline.
Source elements
Source elements (or parsers of formats that provide notion of time, such
as MPEG, as explained above) must place a timestamp in each buffer that
they deliver. The origin of the time used is arbitrary, but it must
match the time delivered in the newsegment event (see below).
However, it is expected that the origin is the origin of the media
stream.
In order to initialize the element time of the rest of the pipeline, a
source element must send a newsegment event before starting to play.
In addition, after seeking, a newsegment event must be sent, because
the timestamp of the next element does not match the element time of the
rest of the pipeline.
Sink elements
If the element is intended to emit samples at a specific time (real time
playing), the element should require a clock, and thus implement the
method set_clock.
In addition, before playing each sample, if the current element time is
less than the timestamp in the sample, it wait until the current time
arrives should call gst_element_wait()
With some schedulers, gst_element_wait()
blocks the pipeline. For instance, if there is one audio sink element
and one video sink element, while the audio element is waiting for a
sample the video element cannot play other sample. This behaviour is
under discussion, and might change in a future release.