audioresample is derived from GstBaseTransform, and one of
GstBaseTransform's traits is that if the derived element does not
produce an output buffer from some input buffer then the first output
buffer after that gets flaged as a discontinuity, whether or not the
buffer actually is discontinuous from the output buffer that preceded
it. When downsampling, the audioresample element requires more than
one input sample for each output sample, and if the ratio of input to
output sample rates is high enough and the input buffers short enough
it can come to pass that the resampler does not receive enough samples
on its input to produce any output. Currently the resampler returns
GST_BASE_TRANSFORM_FLOW_DROPPED from the transform() method in this case,
causing the next buffer to be flagged as a discontinuity. If subsequent
elements in the pipeline reset themselves on disconts, this can cause
clicks and other undesireable behaviour.
Fixes bug #665004.
Remove the _ in front of the endianness prefix.
Remove the _3 postfix for the 24 bits formats.
Add a _32 postfix after the formats that occupy extra space beyond their
natural size.
The result is that the GST_AUDIO_NE() macro can simply append the endianness
after all formats and that we only specify a different sample width when it is
different from the natural size of the sample. This makes things more consistent
and follows the pulseaudio conventions instead of the alsa ones.
Rework the audio caps similar to the video caps. Remove
width/depth/endianness/signed fields and replace with a simple string
format and media type audio/x-raw.
Create a GstAudioInfo and some helper methods to parse caps.
Remove duplicate code from the ringbuffer and replace with audio info.
Use AudioInfo in the base audio filter class.
Port elements to new API.
If the second and next caps structures are a subset of the already existing
transformed caps we can safely skip them because we would transform them to
the same caps again.
Since we calculate timestamps by:
timestamp = t0 + (out samples) / (out rate)
and durations by:
duration = ((out samples) + (processed samples)) / (out rate) - timestamp
if t0 is nonzero, this would simplify to
duration = t0 + (processed samples) / (out rate).
This duration is too large by the amount t0. We should have done:
duration = t0 + ((out samples) + (processed samples)) / (out rate) - timestamp
so that
duration = (processed samples) / (out rate).
Orc is not a hard requirement. Things should still compile and
work without orc, but slow fallback code may be used in this
case. Fix up configure to not error out if orc is not installed
and wrap use of orc profiling in audioresample in #ifdefs.
Fixes#620136 some more.
