When an AudioCallbackDriver starts a fallback SystemClockDriver is running in
its stead. The AudioTrack getting fed data from the input stream of the driver
will get real data while the AudioCallbackDriver is running, and silence while
the fallback driver is running.
If the MediaStreamTrack representing the microphone source is part of a
MediaStream being played by an HTMLMediaElement; and another MediaStreamTrack
representing another source with a lower channel count than the microphone is
part of another MediaStream being played by another HTMLMediaElement; then:
1. We start the graph with the other source's output channel count, and a
fallback driver.
2. Once the audio driver has started, it adds data at a higher channel count
than the graph's to its MediaTrack. The driver switches audio driver to
match the new channel count.
3. The new driver starts with a fallback driver, which adds silence to the
track. Silence has no channel count, so the graph sees only the channel
count of the other source and switches audio driver to match this.
4. Go to 1.
This patch fixes makes us memoize a previously returned max channel count for an
AudioSegment for use when there is only null data (e.g., silence) present in the
segment. This applies to step 3 above, where no audio driver would be switched
because the graph still sees the mic's channel count.
Differential Revision: https://phabricator.services.mozilla.com/D95931
This change looks bigger than it intuitively should. This is because it adds an
input while we wait for the restart.
When we are on the fallback driver, it will only run (from EnsureNextIteration)
if it needs to, i.e., if there are commands or data to process. We open an audio
input that will generate data for the graph, keeping the fallback driver
running.
We could have kept it running in other ways too, but this seemed easier to
regress without noticing -- an active input that sees an error and then
restarting it doesn't work.
Differential Revision: https://phabricator.services.mozilla.com/D94567
This change looks bigger than it intuitively should. This is because it adds an
input while we wait for the restart.
When we are on the fallback driver, it will only run (from EnsureNextIteration)
if it needs to, i.e., if there are commands or data to process. We open an audio
input that will generate data for the graph, keeping the fallback driver
running.
We could have kept it running in other ways too, but this seemed easier to
regress without noticing -- an active input that sees an error and then
restarting it doesn't work.
Differential Revision: https://phabricator.services.mozilla.com/D94567
Allow-list all Python code in tree for use with the black linter, and re-format all code in-tree accordingly.
To produce this patch I did all of the following:
1. Make changes to tools/lint/black.yml to remove include: stanza and update list of source extensions.
2. Run ./mach lint --linter black --fix
3. Make some ad-hoc manual updates to python/mozbuild/mozbuild/test/configure/test_configure.py -- it has some hard-coded line numbers that the reformat breaks.
4. Make some ad-hoc manual updates to `testing/marionette/client/setup.py`, `testing/marionette/harness/setup.py`, and `testing/firefox-ui/harness/setup.py`, which have hard-coded regexes that break after the reformat.
5. Add a set of exclusions to black.yml. These will be deleted in a follow-up bug (1672023).
# ignore-this-changeset
Differential Revision: https://phabricator.services.mozilla.com/D94045
This renames the thread and identifiers derived from the thread's name. This is
to avoid ambiguity over if the thread relates to the MediaController class,
which it does not.
Differential Revision: https://phabricator.services.mozilla.com/D93806
Allow-list all Python code in tree for use with the black linter, and re-format all code in-tree accordingly.
To produce this patch I did all of the following:
1. Make changes to tools/lint/black.yml to remove include: stanza and update list of source extensions.
2. Run ./mach lint --linter black --fix
3. Make some ad-hoc manual updates to python/mozbuild/mozbuild/test/configure/test_configure.py -- it has some hard-coded line numbers that the reformat breaks.
4. Make some ad-hoc manual updates to `testing/marionette/client/setup.py`, `testing/marionette/harness/setup.py`, and `testing/firefox-ui/harness/setup.py`, which have hard-coded regexes that break after the reformat.
5. Add a set of exclusions to black.yml. These will be deleted in a follow-up bug (1672023).
# ignore-this-changeset
Differential Revision: https://phabricator.services.mozilla.com/D94045
Allow-list all Python code in tree for use with the black linter, and re-format all code in-tree accordingly.
To produce this patch I did all of the following:
1. Make changes to tools/lint/black.yml to remove include: stanza and update list of source extensions.
2. Run ./mach lint --linter black --fix
3. Make some ad-hoc manual updates to python/mozbuild/mozbuild/test/configure/test_configure.py -- it has some hard-coded line numbers that the reformat breaks.
4. Add a set of exclusions to black.yml. These will be deleted in a follow-up bug (1672023).
# ignore-this-changeset
Differential Revision: https://phabricator.services.mozilla.com/D94045
Without this patch, the FrameEncode gtest encoder won't encode the last bit of
data, making the total duration 20ms too short.
When passing EOS we encode the lookahead worth of silence, so this patch also
accounts for that.
Differential Revision: https://phabricator.services.mozilla.com/D91957
Certain logic, like AudioOutputFramesPerPacket(), is hinged off whether a
resampler exists. The resampler is destroyed when encoding is completed, making
such logic flawed from that point. To avoid this potential footgun we can decide
the output rate at construction time, since the input rate is known then.
Differential Revision: https://phabricator.services.mozilla.com/D91953
Without this patch there was a gap between the default-ctor and when real values
got set. If setting a member was forgotten, it would have needed an audit to be
found. With this patch the compiler will make sure all values have been
explicitly handed to the ctor.
mFrameData (nsTArray) becomes Refcountable to allow VP8TrackEncoder to extend
the duration of an EncodedFrame by copying the last frame's ref and constructing
a new EncodedFrame with a longer duration than the last one's.
Differential Revision: https://phabricator.services.mozilla.com/D91724
This was originally handled by EbmlComposer. Since bug 1014393 this was handled
by MediaEncoder. By doing it in OpusTrackEncoder we can avoid reading hardcoded
fields in the opus metadata to get the codec delay value.
Differential Revision: https://phabricator.services.mozilla.com/D91723
Fix up includes so AnnexB.h and TestMediaDataEncoder.cpp don't rely on unified
build order. Reformat include lists to match style guide. Rework #include guard
on AnnexB.h to reflect style guide.
Differential Revision: https://phabricator.services.mozilla.com/D91825
One second is a bit short given that starting bluetooth input devices takes
several seconds on some platforms. Bug 1586370 may have worsened this since we
now process silence while waiting for an input to give us data.
Differential Revision: https://phabricator.services.mozilla.com/D91140
This has busted the MingW build, where webrtc is not supported.
AudioInputProcessing relies on upstream webrtc for audio processing: aec, etc.
Differential Revision: https://phabricator.services.mozilla.com/D90852
This patch simplifies the logic by reducing the number of paths, from two to
one method that changes the correction. This method now weighs the calculated
correction by 60% and the previous correction by 40% when setting the new value,
to provide a negative feedback loop that stabilizes around the desired amount of
buffering.
This leads to a smoother output with less noticable changes in the correction
value, while still reaching the desired amount of buffering quickly.
Tests are updated with new expectations accordingly.
Differential Revision: https://phabricator.services.mozilla.com/D89776
With this patch, there is a fake audio thread present on a MockCubeb context as
soon as one MockCubebStream is running under that context. When the last running
MockCubebStream is stopped, the fake audio thread is joined and unset.
This adds a tad bit of complexity but results in zero unwanted drift between
MockCubebStreams under the same MockCubeb context. This is essential for stable
CrossGraphTrack tests.
A side effect of this is that the drift factor of a MockCubebStream does not
affect the interval at which data is processed, but rather the amount of data
processed each interval.
This patch also allows us to process data with virtually no wait time between
iterations (as opposed to wall-time 10ms-waits), for (much) speedier tests.
Differential Revision: https://phabricator.services.mozilla.com/D89772
Andreas Pehrson
John Lin
Sylvestre Ledru
Butkovits Atila
Andi-Bogdan Postelnicu
Csoregi Natalia
Ricky Stewart
Bryce Seager van Dyk
Bogdan Tara
Dorel Luca
Matthew Gregan
Dan Minor
Simon Giesecke
Nico Grunbaum