The new interface uses optionals instead of default values, and only values that are actually used are included. To make it easy to add/remove stats in the future, the struct itself is copied around on all layers, instead of copying the values one by one. This CL also fixes a bug which caused several APM statistics to get stuck at a fixed level when there are no more receive streams (after some period where there were receive streams). Since APM doesn't know this happens, an argument was added to the GetStats call to pass this information down to APM.
Bug: webrtc:8563, b/67926135
Change-Id: I96cc008353355bb520c4523f5c5379860f73ee24
Reviewed-on: https://webrtc-review.googlesource.com/25621
Commit-Queue: Ivo Creusen <ivoc@webrtc.org>
Reviewed-by: Henrik Boström <hbos@webrtc.org>
Reviewed-by: Karl Wiberg <kwiberg@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20877}
This test produces a consistent stream of false positive alerts, and I have been unable to make it more robust, despite several attempts. It also has never managed to catch a real regression, so I think it is better to remove it.
Bug: chromium:788318
Change-Id: I7e9731834f67af1ef2fa15a655e620bd64a4cfde
Reviewed-on: https://webrtc-review.googlesource.com/25824
Commit-Queue: Ivo Creusen <ivoc@webrtc.org>
Reviewed-by: Alex Loiko <aleloi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20874}
The audio processing module reports the metrics 'echo return loss'
and 'echo return loss enhancement' for AEC3 via the old, soon to be
replaced, GetStatistics fuction.
Bug: webrtc:8533
Change-Id: I6b2286b5cdf8f20ebf14f82f1180f6bfb6c00c68
Reviewed-on: https://webrtc-review.googlesource.com/25642
Commit-Queue: Gustaf Ullberg <gustaf@webrtc.org>
Reviewed-by: Henrik Lundin <henrik.lundin@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20869}
The APM-QA tool produces clean-speech + noise + echo mixes with the
--additive_noise_tracks_path, --test_data_generators,
--echo_path_simulator flags. From this CL, it automatically produces
compressed Numpy annotations for the mixes. Annotations are placed in
the same folder as the mixes with name '${basename}-annotations.npz'.
TBR=alessiob@webrtc.org
NOTRY=True
Bug: webrtc:7494
Change-Id: I71941a4283594ef813de3ed65be31623bce5d7de
Reviewed-on: https://webrtc-review.googlesource.com/24960
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Reviewed-by: Alex Loiko <aleloi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20844}
The audio processing module reports the metrics 'echo return loss'
and 'echo return loss enhancement' for AEC3.
Bug: webrtc:8533
Change-Id: I166c504adf013d6cb5d6d3c9717d0622c3454bb7
Reviewed-on: https://webrtc-review.googlesource.com/24880
Commit-Queue: Gustaf Ullberg <gustaf@webrtc.org>
Reviewed-by: Ivo Creusen <ivoc@webrtc.org>
Reviewed-by: Per Åhgren <peah@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20835}
Specifically, I'm moving
histogram_percentile_counter.h
mathutils.h
mod_ops.h
moving_max_counter.h
They shouldn't be part of the API, and moving them to an appropriate
subdirectory of rtc_base/ is a good way to keep track of that.
BUG=webrtc:8445
Change-Id: I78a999984a27ef935be2d7c3136475d5f209adda
Reviewed-on: https://webrtc-review.googlesource.com/20870
Commit-Queue: Karl Wiberg <kwiberg@webrtc.org>
Reviewed-by: Danil Chapovalov <danilchap@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20832}
Specifically, I'm moving
safe_compare.h
safe_conversions.h
safe_minmax.h
They shouldn't be part of the API, and moving them to an appropriate
subdirectory of rtc_base/ is a good way to keep track of that.
BUG=webrtc:8445
Change-Id: I458531aeb30bcf4291c4bec3bf22a2fffbf054ff
Reviewed-on: https://webrtc-review.googlesource.com/20860
Commit-Queue: Karl Wiberg <kwiberg@webrtc.org>
Reviewed-by: Danil Chapovalov <danilchap@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20829}
There is now an 'ExternalVad' class in the AnnotationsExtractor. The
Extractor takes an extra list of these in addition to the other
VADs. The external VAD runs an external program to generate the
annotations. Annotations are loaded and saved to a compressed Numpy format.
Also made a small fix to name a mixed file in a way so that files will not
be overwritten.
Also did some minor changes to the unittests.
TBR=alessiob@webrtc.org
Bug: webrtc:7494
Change-Id: I7816b04466be16cd635ac6ceab18cd7aad5325a4
Reviewed-on: https://webrtc-review.googlesource.com/23623
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Reviewed-by: Alex Loiko <aleloi@webrtc.org>
Reviewed-by: Alessio Bazzica <alessiob@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20819}
This new parallel GetStatistics function uses Optionals to indicate if stats are valid or not, and no longer relies on default values. It also takes an argument to indicate if receive streams are present, and if not several stats will not be set.
Bug: b/67926135
Change-Id: I175de1c65c414bea6ec9ca8b0b16f07cb2308d9f
Reviewed-on: https://webrtc-review.googlesource.com/17942
Commit-Queue: Ivo Creusen <ivoc@webrtc.org>
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Reviewed-by: Henrik Boström <hbos@webrtc.org>
Reviewed-by: Henrik Lundin <henrik.lundin@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20789}
Changes places where we explicitly construct an Optional to instead use
nullopt or the requisite value type only.
This CL was uploaded by git cl split.
R=henrik.lundin@webrtc.org
Bug: None
Change-Id: I733a83f702fe11884d229a1713cfac952727bde8
Reviewed-on: https://webrtc-review.googlesource.com/23601
Commit-Queue: Oskar Sundbom <ossu@webrtc.org>
Reviewed-by: Henrik Lundin <henrik.lundin@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20786}
This CL refactors the delay estimator in AEC3.
Furthermore, it adds:
1. Allow for a customized delay estimator behavior to simplify
development.
2. Exposes that behavior to clear configuration settings.
3. Adds logging of the delay range supported by the delay
estimator.
Bug: webrtc:8519
Change-Id: I1764a090519a78b021b2e7de565c52a6c02c848e
Reviewed-on: https://webrtc-review.googlesource.com/21166
Commit-Queue: Per Åhgren <peah@webrtc.org>
Reviewed-by: Gustaf Ullberg <gustaf@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20733}
This makes it easier to import cleanly downstream, and makes it
a lot easier to reason about.
Bug: webrtc:6828
Change-Id: I953e129de73053f8619333fe7e318b36e3a1fffa
Reviewed-on: https://webrtc-review.googlesource.com/22722
Commit-Queue: Patrik Höglund <phoglund@webrtc.org>
Reviewed-by: Henrik Lundin <henrik.lundin@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20727}
Conditional visibility is complex to maintain and it is not well
supported by other build systems.
This CL removes it and falls back on the more relaxed visibility value
("*" in this case).
It is not a problem because the targets that are using conditional
visibility are all marked as "testonly" and this is probably enough to
keep the build graph clean.
Bug: None
Change-Id: I2d2b5ac9a02d08c2863950116db455976ee1459c
No-Try: True
Reviewed-on: https://webrtc-review.googlesource.com/14902
Commit-Queue: Mirko Bonadei <mbonadei@webrtc.org>
Reviewed-by: Patrik Höglund <phoglund@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20658}
Between patch set 4 and patch set 5 in
https://codereview.webrtc.org/2865113002/, a line consisting of a
single 'std::move(task);' was added. The reason we will never know,
because the author will not tell. The superfluous line would have gone
unnoticed except for occasional raised eyebrows of casual code
readers.
The Visual Studio compiler now sees lines that have no effect. Which
was announced to the world in the tweet
https://twitter.com/StephanTLavavej/status/924011366943354880
achieving 27 likes and 6 retweets.
Bug: webrtc:8463
Change-Id: Iac49bc4153254b6cfe99f609da28eb4f43ff765e
Reviewed-on: https://webrtc-review.googlesource.com/21324
Reviewed-by: Sam Zackrisson <saza@webrtc.org>
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20616}
An energy value is calculated by summing squares of processed audio
samples. The expression 'out*out >> 6' could overflow. In this CL we
change it to 'out*(out>>6) + out*(out*(out%(1<<6))>>6)'.
The which is verified and proven to be equal, but doesn't
overflow. The change also passes our change-detection tests in
GainControlBitExactnessTest.*
We verified with Godbolt that the modulo and divisions are converted
into branch-free bitwise operations.
NOTRY=True # changing comment, tests just passed.
Bug: chromium:780638, chromium:780376
Change-Id: I415535193433a2fbc275c643fb4e4026ba3e0bdd
Reviewed-on: https://webrtc-review.googlesource.com/20867
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Reviewed-by: Sam Zackrisson <saza@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20589}
Added possibility to extract audio_processing VAD annotations in the Quality Assessment tool.
Annotations are extracted into compressed Numpy 'annotations.npz' files.
Annotations contain information about VAD, speech level, speech probabilities etc.
TBR=alessiob@webrtc.org
Bug: webrtc:7494
Change-Id: I0e54bb67132ae4e180f89959b8bca3ea7f259458
Reviewed-on: https://webrtc-review.googlesource.com/17840
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Reviewed-by: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Alex Loiko <aleloi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20581}
The old value was 170, but experiments have shown that 70 is better.
This will let the AGC reduce the gain further when input clipping is
detected. The effect should be less clipping, but sometimes slightly
lower signals.
In Chrome, the value 70 has already been used since June (see
https://codereview.chromium.org/2928133002).
Bug: webrtc:6622, chromium:672476
Change-Id: Ie5a60bb875eef71f303b28e096b22a8cd4b449d4
Reviewed-on: https://webrtc-review.googlesource.com/20222
Reviewed-by: Per Åhgren <peah@webrtc.org>
Commit-Queue: Henrik Lundin <henrik.lundin@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20563}
This CL replaces one 'int32_t' with 'uint32_t'. The value is a
non-negative energy, and the number of leading zeros is
computed. During computation, a shift can cause it to overflow.
Issue was found by the Audio Processing fuzzer.
Bug: chromium:778939, chromium:778921, chromium:778919
Change-Id: I3d7e0b547e6b0edcd9995903517ea851142a08c1
Reviewed-on: https://webrtc-review.googlesource.com/16433
Reviewed-by: Sam Zackrisson <saza@webrtc.org>
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20470}
This CL lowers the threshold for delay change detection in AEC3.
This makes the delay decisions more stable.
TBR=gustaf@webrtc.org
Bug: chromium:778396,webrtc:8451
Change-Id: I8b015455399d696172b7c0beb033caf508f426e9
Reviewed-on: https://webrtc-review.googlesource.com/15541
Reviewed-by: Per Åhgren <peah@webrtc.org>
Commit-Queue: Per Åhgren <peah@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20433}
Some death tests for AEC3 cause memory leaks on trybots. This CL
temporarily disables BlockProcessor.VerifyRenderBlockSizeCheck.
Bug: webrtc:8449,webrtc:6985
Change-Id: I2900a73f7c7d5bf0e8b58a20f9a40bd5d654629a
Reviewed-on: https://webrtc-review.googlesource.com/15500
Commit-Queue: Gustaf Ullberg <gustaf@webrtc.org>
Reviewed-by: Oleh Prypin <oprypin@webrtc.org>
Reviewed-by: Per Åhgren <peah@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20431}
This CL replaces 5 left shifts where the shifted value may be
negative. The shifts are replaced with equivalent multiplications.
Bug: chromium:777231, chromium:776719, chromium:776624, chromium:776286
Change-Id: Ifb27d5506eac779e60f238432bdf9e4bc5b2da4c
Reviewed-on: https://webrtc-review.googlesource.com/14800
Reviewed-by: Henrik Lundin <henrik.lundin@webrtc.org>
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20430}
This CL balances the NLP tradeoff in AEC3 to properly handle the cases
when the echo path is so strong that it saturates the echo and when it
is so weak that the echo is very low compared to nearend.
Bug: webrtc:8411, webrtc:8412, chromium:775653
Change-Id: I5aff74dfadd51cac1ce71b1cb935d68a5be6918d
Reviewed-on: https://webrtc-review.googlesource.com/14120
Commit-Queue: Per Åhgren <peah@webrtc.org>
Reviewed-by: Per Åhgren <peah@webrtc.org>
Reviewed-by: Gustaf Ullberg <gustaf@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20418}
In the legacy C part of AGC, an audio level 'cur_level' is represented as
(1+frac) * 2^(31 - zeros)
The 'zeros' exponent part is used for looking up a gain value in a
table, and 'frac' is used for interpolating between two nearby table
values. Code snippet below:
zeros = WebRtcSpl_NormU32((uint32_t)cur_level);
tmp32 = (cur_level << zeros) & 0x7FFFFFFF;
frac = (int16_t)(tmp32 >> 19);
In the second line, 'cur_level' is shifted upwards so that the leading
bit is '1', after which the leading bit is cleared. The result is
'frac' in Q31.
The compiler type of 'cur_level << zeros' is 'int32_t'. This is a
fuzzer error 'Left shift cannot be represented in int32_t',
because the leading sign bit is 1. This CL changes the compiler type to
uint32_t.
Bug: chromium:776286
Change-Id: Ie29552b75e690057bd76fc88e747841b531e3802
Reviewed-on: https://webrtc-review.googlesource.com/14841
Reviewed-by: Alessio Bazzica <alessiob@webrtc.org>
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20405}
Alternative VAD based on the existing one in WebRTC.
It is used to extract VAD annotations in APM-QA.
TBR=
Bug: webrtc:7494
Change-Id: I6af412742f804631ad4f3ba3ccf71a30d74de984
Reviewed-on: https://webrtc-review.googlesource.com/14553
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Alessio Bazzica <alessiob@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20404}
This CL includes the patch from oprypin@webrtc.org, which is also applied
to the MIPS code (also affected), and the protobuf for ApmTest.Process
(audio_processing_unittest.cc), which used when WEBRTC_AUDIOPROC_FIXED_PROFILE
is set.
This change has been tested on mobile platforms.
Bug: webrtc:8200
Change-Id: Ic50a5ab57c16551397756b1fb473e1067b8e7ece
Reviewed-on: https://webrtc-review.googlesource.com/10811
Reviewed-by: Per Åhgren <peah@webrtc.org>
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20394}
The 'parametricNoise' field is never initialized in the
'WebRtcNs_InitCore' function that initializes a 'NoiseSuppressionC'
struct.
This leads to use of unititialized value, which may affect the audio
output and result of the noise suppressor.
The issue was found by the Chrome fuzzer:
https://clusterfuzz.com/v2/testcase-detail/4749034115039232
Bug: chromium:776673
Change-Id: I1c3fd80cff178f2d5917064ad07f88c7b9a29e7d
Reviewed-on: https://webrtc-review.googlesource.com/14556
Reviewed-by: Ivo Creusen <ivoc@webrtc.org>
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20388}
AddRef() now returns void, and Release() returns an enum
RefCountReleaseStatus, to indicate whether or not this Release
call implied deletion.
Bug: webrtc:8270
Change-Id: If2fb77f26118b61751b51c856af187c72112c630
Reviewed-on: https://webrtc-review.googlesource.com/3320
Commit-Queue: Niels Moller <nisse@webrtc.org>
Reviewed-by: Karl Wiberg <kwiberg@webrtc.org>
Reviewed-by: Taylor Brandstetter <deadbeef@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20366}
The struct containing the config for AEC3 is removed from
AudioProcessing::Config and is put in a new struct called
EchoCanceller3Config.
AEC3 should no longer be activated through
AudioProcessing::ApplyConfig. Instead an EchoCanceller3Factory
can be injected at AudioProcessing creation.
Bug: webrtc:8346
Change-Id: I27e3592e675eec3632a60c45d9e0d12514c2c567
Reviewed-on: https://webrtc-review.googlesource.com/11420
Reviewed-by: Per Åhgren <peah@webrtc.org>
Reviewed-by: Henrik Lundin <henrik.lundin@webrtc.org>
Commit-Queue: Gustaf Ullberg <gustaf@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20342}
Extract and save some simple annotations for the clean speech input.
The annotations are estimated level, VAD (assuming clean speech) and speech level.
TBR=
Bug: webrtc:7494
Change-Id: Id73358e228fac721a77fc8a61a3474a5d52bdc84
Reviewed-on: https://webrtc-review.googlesource.com/12321
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Alessio Bazzica <alessiob@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20327}
We are using <math.h>, not <cmath>. While the latter defines additional
overloads for abs(), including abs(float), they are not guaranteed to be
available in <math.h>.
libc++ ships its own math.h with the additional overloads, and libstdc++ (v6
or later) has a math.h that includes <cmath>, but this is not always
expected to work: for example, GCC 5.x's libstdc++ does not have these
additional overloads and causes the build to fail.
Just use fabsf() from the C standard library directly, as it achieves the
same thing in a more portable fashion.
Bug: None
Change-Id: I805728269b35051edb54126e204eccd2706e3a92
Reviewed-on: https://webrtc-review.googlesource.com/11460
Reviewed-by: Henrik Lundin <henrik.lundin@webrtc.org>
Commit-Queue: Raphael Kubo da Costa (rakuco) <raphael.kubo.da.costa@intel.com>
Cr-Commit-Position: refs/heads/master@{#20325}
This CL changes the filter delay detection to rely on the largest peak
while the correctness of the filter is changed to be based on the
performance achieved by the filter.
Bug: webrtc:8397,chromium:774867
Change-Id: I70c953815192478f9a8e0da9f2b8fd9edac3f481
Reviewed-on: https://webrtc-review.googlesource.com/10803
Commit-Queue: Per Åhgren <peah@webrtc.org>
Reviewed-by: Gustaf Ullberg <gustaf@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20321}
This CL changes the AEC3 behavior to be more transparent when there
is uncertainty about the amount of echo in the microphone signal.
Bug: webrtc:8398, chromium:774868
Change-Id: I88e681f8decd892f44397b753df371a1c4b90af0
Reviewed-on: https://webrtc-review.googlesource.com/10801
Reviewed-by: Gustaf Ullberg <gustaf@webrtc.org>
Commit-Queue: Per Åhgren <peah@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20319}
Echo Control is enabled in capture_nonlocked_ when injected.
Renamed echo_canceller3_enabled to echo_controller_enabled.
Bug: webrtc:8346
Change-Id: Icf441f07ce64719358841544da7579feeb7cfdbb
Reviewed-on: https://webrtc-review.googlesource.com/10808
Commit-Queue: Gustaf Ullberg <gustaf@webrtc.org>
Reviewed-by: Per Åhgren <peah@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20311}
Added EchoCanceller3Factory that implements EchoControlFactory and can
be used for injecting EchoCanceller3 into Audio Processing Module.
Renamed InitializeEchoCanceller3 to InitializeEchoController.
Bug: webrtc:8346
Change-Id: I47078da6a49aca1ee41f6a9d5b7b8e91bb5c11a3
Reviewed-on: https://webrtc-review.googlesource.com/9220
Commit-Queue: Gustaf Ullberg <gustaf@webrtc.org>
Reviewed-by: Per Åhgren <peah@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20299}
The field trial effects two things: after a frame length change the IAT
histogram is scaled to prevent an immediate change in target buffer
level. Also, the peak history in the delay peak detector is cleared,
because the size of the peaks is stored in number of packets (which
will be incorrect after a frame length change).
Bug: webrtc:8381
Change-Id: I214b990f6e5959b655b6542884a7f75da181a0d8
Reviewed-on: https://webrtc-review.googlesource.com/8101
Reviewed-by: Henrik Lundin <henrik.lundin@webrtc.org>
Commit-Queue: Ivo Creusen <ivoc@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20284}
In preparation of coming CLs that will add an AGC interface to make the
gain controller injectable.
This CL simplifies AGC2 (dummy sub-module of audioproc_f) since it only
implements the fixed digital mode with hard-clipping - i.e., no limiter
is used.
The AGC2 config now includes the fixed gain to apply and audioproc_f
has been adapted accordingly.
Finally, this CL slightly simplifies the AGC2 integration into APM.
This CL is a continuation of https://codereview.webrtc.org/2995043002/
Bug: webrtc:7494
Change-Id: I3d554ea4dc6208928352059feb14987edabf14c7
Reviewed-on: https://webrtc-review.googlesource.com/4661
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Per Åhgren <peah@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20278}
The factory for EchoControl is changed from an rtc::Callback1 to an
interface. This avoids using rtc::Callback1 outside of WebRTC.
This also makes the EchoControl factory more similar to other
factories in the code base.
Bug: webrtc:8345
Change-Id: Ie61b9416ed771f8c756326736d17e339eb768469
Reviewed-on: https://webrtc-review.googlesource.com/8900
Reviewed-by: Per Åhgren <peah@webrtc.org>
Commit-Queue: Gustaf Ullberg <gustaf@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20272}
For each single experiment, a URL is defined by adding a specific anchor.
A URL can be copied using the button beneath the score of the experiment
one would like to share.
This CL also includes a few optimizations and fixes:
- JS and CSS are minified
- Dialog close event listener added, this fixes a small bug preventing
the play out audio to stop when pressing ESC instead of using the close
button
- Snackbar notifications added
- Simple unit test for the export module
BUG=webrtc:7218
Change-Id: Iad00ce69094a5968ee0520d105d59656cfafa4e2
TBR=
Change-Id: Iad00ce69094a5968ee0520d105d59656cfafa4e2
Reviewed-on: https://webrtc-review.googlesource.com/7960
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Alessio Bazzica <alessiob@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#20266}
