This CL does *not* change the behavior of the AGC2 adaptive digital
controller - bitexactness verified with audioproc_f on a collection of
AEC dumps and Wav files (42 recordings in total).
Tested: compiled Chrome with this patch and made an appr.tc test call
Bug: webrtc:7494
Change-Id: Ia8a9f6fbc3a3459b888a2eed87e108f0d39cfe99
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/233520
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Sam Zackrisson <saza@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35140}
Instead of using two different headroom parameters, namely
`kHeadroomDbfs` and `kSaturationProtectorExtraHeadroomDb`, only use
the former that now also accounts for the deleted one - i.e., it equals
the sum of the two headrooms. In this way, tuning AGC2 will be easier.
This CL does *not* change the behavior of the AGC2 adaptive digital
controller - bitexactness verified with audioproc_f on a collection of
AEC dumps and Wav files (42 recordings in total).
The unit tests changes in agc2/saturation_protector_unittest.cc are
required since `extra_headroom_db` is removed and the changes in
agc2/adaptive_digital_gain_applier_unittest.cc are required because
`AdaptiveDigitalGainApplier` depends on `kHeadroomDbfs` which has been
updated as stated above.
Bug: webrtc:7494
Change-Id: I0a2a710bbede0caa53938090a004d185fdefaeb9
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/232905
Reviewed-by: Per Åhgren <peah@webrtc.org>
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35109}
Add the option to run the adaptive digital controller of AGC2 without
side-effects - i.e., no gain applied.
Tested: adapation verified during a video call in chromium
Bug: webrtc:7494
Change-Id: I4776f6012907d76a17a3bca89991da97dc38657f
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/215964
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Per Åhgren <peah@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#33875}
- Bug fix: the desired initial gain quickly dropped to 0 dB hence
starting a call with a too low level
- New tuning to make AGC2 more robust to VAD mistakes
- Smarter max gain increase speed: to deal with an increased threshold
of adjacent speech frames, the gain applier temporarily allows a
faster gain increase to deal with a longer time spent waiting for
enough speech frames in a row to be observed
- Saturation protector isolated from `AdaptiveModeLevelEstimator` to
simplify the unit tests for the latter (non bit-exact change)
- AGC2 adaptive digital config: unnecessary params deprecated
- Code readability improvements
- Data dumps clean-up and better naming
Bug: webrtc:7494
Change-Id: I4e36059bdf2566cc2a7e1a7e95b7430ba9ae9844
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/215140
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Jesus de Vicente Pena <devicentepena@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#33736}
The current noise level estimator has a bug due to which the estimated
level decays to the lower bound in a few seconds when speech is observed.
Instead of fixing the current implementation, which is based on a
stationarity classifier, an alternative, lightweight, noise floor
estimator has been added and tuned for AGC2.
Tested on several AEC dumps including HW mute, music and fast talking.
Bug: webrtc:7494
Change-Id: Iae4cff9fc955a716878f830957e893cd5bc59446
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/214133
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Per Åhgren <peah@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#33733}
Make the digital adaptive gain applier more robust to VAD false
positives. Achieved by allowing a gain increase only if enough adjacent
speech frames are observed.
Tested:
- Bit-exactness verified with audioproc_f
- If `kDefaultDigitalGainApplierAdjacentSpeechFramesThreshold` == 2
then not bit-exact
Bug: webrtc:7494
Change-Id: I3bab5a449aaf0ef1a64b671b413ba2ddb4688cd2
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/186042
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Ivo Creusen <ivoc@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32263}
This is the last CL needed to add a new `AdaptiveModeLevelEstimator`
feature that makes AGC2 more robus to VAD mistakes: the level estimator
discards estimation updates when too few consecutive speech frames are
observed.
This CL adds a second state property to hold temporary updates and a
counter for consecutive speech frames. When enough speech frames are
observed, the reliable state is updated; otherwise, the temporary state
is discarded.
The default for `AdaptiveModeLevelEstimator::min_consecutive_speech_frames_`
is 1, which means that the new feature is disabled.
Tested:
- Bit-exactness verified with audioproc_f
- Not bit-exact if `min_consecutive_speech_frames_` set to 10
Bug: webrtc:7494
No-Try: True
Change-Id: I0daa00e90c27c418c00baec39fb8eacd26eed858
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/185125
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Per Åhgren <peah@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32250}
Feature added to gain robustness to occasional VAD speech probability
spikes. In such a case, the attack process reduces the chance that the
smoothed values are greater than the speech threshold.
Bug: webrtc:7494
Change-Id: I6babe5afe30ea3dea021181a19d86bb74b33a98c
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/185046
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Gustaf Ullberg <gustaf@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32198}
Changing VAD (voice activity detector) confidence threshold from 40%
to 90%. The proportion of samples classified as speech drops to ca 80%
of what it was when the threshold was 40%. Therefore,
kFullBufferSizeMs has to be increased by 1.0/0.8. We increase it from
1600ms to 2000ms.
TESTED = Did run the new and old configs on AEC dumps. With one minute
of kitchen noise, the new tuning boosted the noise by 3-4 db less.
Bug: chromium:913430
Change-Id: I4a2ebb6d1d309c6c20dd23c3685818b1b5ad4a66
Reviewed-on: https://webrtc-review.googlesource.com/c/113806
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Reviewed-by: Alessio Bazzica <alessiob@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#25950}
This CL is the result of running include-what-you-use tool on part
of the code base (audio target and dependencies) plus manual fixes.
bug: webrtc:8311
Change-Id: I277d281ce943c3ecc1bd45fd8d83055931743604
Reviewed-on: https://webrtc-review.googlesource.com/c/106280
Commit-Queue: Yves Gerey <yvesg@google.com>
Reviewed-by: Mirko Bonadei <mbonadei@webrtc.org>
Reviewed-by: Patrik Höglund <phoglund@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#25311}
The AdaptiveAgc often boosts the signal outside of Float S16 range. It
is expected, which is why we have a limiter after it in the process
chain. But it turns out that this happens regularly even for simple
input examples. The output signal peaks can be as high as +4 dBFs for a
single speaker example (which should be easy). It leads to excessive
gain modulation by the limiter.
This CL is a new tuning designed to produce a safer gain. After this,
we shouldn't hit the saturation region of the limiter as often. But we
will still maintain a high gain.
We have a 'configurable kill-switch': the settings can be changed via
field trials WebRTC-Audio-Agc2Force(Initial|Extra)SaturationMargin.
Bug: webrtc:7494, chromium:892043
Change-Id: I5014377050c74c32ae8998282991141eae31cf58
Reviewed-on: https://webrtc-review.googlesource.com/c/102922
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Reviewed-by: Sam Zackrisson <saza@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#25006}
Hypothetical scenario: short weak speech at start of call, then high
noise. The digital adaptive AGC2 would pick a high gain, and then
continue to apply it on the noise. Unless the noise is detected by the
noise estimator, the gain would never be reduced.
This CL addresses the issue by sending limiter gain info to the
adaptive digital AGC2.
Bug: webrtc:7494
Change-Id: Idf5c2686af0f5e5bad981d39a95b8efc9ffb9d64
Reviewed-on: https://webrtc-review.googlesource.com/102641
Reviewed-by: Sam Zackrisson <saza@webrtc.org>
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#24922}
This CL does the following:
1. Adds a new AdaptiveModeLevelEstimatorAgc implementation of the Agc
interface. The new implementation differs from webrtc::Agc by
1. using the AGC2 speech level estimator in
GetRmsErrorDb. webrtc::Agc implements its own with help of
webrtc::LoudnessHistogram.
2. Doesn't forget its past at every GetRmsErrorDb call.
2. Makes AgcManagerDirect use AdaptiveModeLevelEstimatorAgc instead of
webrtc::Agc if the use_agc2_level_estimation flag is set.
Bug: webrtc:7494
Change-Id: I8df3f52e322d433eb5ce5297f4236af2f1877b04
Reviewed-on: https://webrtc-review.googlesource.com/86603
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Reviewed-by: Sam Zackrisson <saza@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#23875}
* Move 'VadWithLevel' to AGC2 where it belongs.
* Remove the vectors from VadWithLevel. They were there to make it work
with modules/audio_processing/vad, which we don't need any longer.
* Remove the vector handling from AGC2. It was spread out across
AdaptiveDigitalGainApplier, AdaptiveAGC and their unit tests.
* Hack the RNN VAD into VadWithLevel. The main issue is the resampling.
Bug: webrtc:9076
Change-Id: I13056c985d0ec41269735150caf4aaeb6ff9281e
Reviewed-on: https://webrtc-review.googlesource.com/77364
Reviewed-by: Sam Zackrisson <saza@webrtc.org>
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#23688}
Running clang-format with chromium's style guide.
The goal is n-fold:
* providing consistency and readability (that's what code guidelines are for)
* preventing noise with presubmit checks and git cl format
* building on the previous point: making it easier to automatically fix format issues
* you name it
Please consider using git-hyper-blame to ignore this commit.
Bug: webrtc:9340
Change-Id: I694567c4cdf8cee2860958cfe82bfaf25848bb87
Reviewed-on: https://webrtc-review.googlesource.com/81185
Reviewed-by: Patrik Höglund <phoglund@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#23660}
This is a kitchen-sink header, some pieces should be moved to
byteorder.h, the rest likely deleted.
Delete most includes of basictypes.h. In leaf headers,
include stddef.h and stdint.h explicitly where needed.
Bug: webrtc:6853
Change-Id: Ibc809936a8f94d418e4eb650da1e89c1b9142073
Reviewed-on: https://webrtc-review.googlesource.com/77721
Commit-Queue: Niels Moller <nisse@webrtc.org>
Reviewed-by: Fredrik Solenberg <solenberg@webrtc.org>
Reviewed-by: Karl Wiberg <kwiberg@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#23333}
If the adaptive gain is too low, we raise it slowly and only during
speech.
The CL gives better behavior at the start of a call. If the gain is too
high, the fixed-digital limits it. The gain is also quickly reduced by
the AdaptiveGainApplier.
Bug: webrtc:7494
Change-Id: I683f1e3e463cddec2d91f6c7f15c73e744430034
Reviewed-on: https://webrtc-review.googlesource.com/71484
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Reviewed-by: Alessio Bazzica <alessiob@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#23053}
AGC2 component that computes and applies the digital gain.
The gain is computed from an estimated speech and noise level.
This component decides how fast the gain can change and what it
should be.
Bug: webrtc:7494
Change-Id: If55b6e5c765f958e433730cd9e3b2b93c14a7910
Reviewed-on: https://webrtc-review.googlesource.com/64985
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Reviewed-by: Alessio Bazzica <alessiob@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#22741}
Another submodule of the Automatic Gain Controller 2. It refines the
biased estimate of the Adaptive Mode Level Estimator. It works by
generating a delayed stream of peak levels. The delayed peaks are
compared to the level estimate.
Bug: webrtc:7494
Change-Id: If4c2c19088d1ca73fb93511dad4e1c8ccabcaf03
Reviewed-on: https://webrtc-review.googlesource.com/65461
Reviewed-by: Ivo Creusen <ivoc@webrtc.org>
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#22732}
The level estimator (AdaptiveModeLevelEstimator) produces a biased
estimate of the speech level. In our model, we use another module
(the SaturationProtector) to compute the bias. This CL contains the
estimator and a stub of the saturation protector.
Bug: webrtc:7494
Change-Id: I0df736d0346063f544fa680b4cc84177ea548545
Reviewed-on: https://webrtc-review.googlesource.com/64820
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Reviewed-by: Ivo Creusen <ivoc@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#22641}
This CL adds the GainCurveApplier (GCA). It owns a
FixedDigitalLevelEstimator (LE) and an InterpolatedGainCurve
(IGC). The GCA uses the LE to compute the input signal level, looks up
a gain from IGC and applies it on the signal.
The other IGC and LE submodules were added in previous CLs [1] and
[2].
This CL also turns on AGC2 in the APM fuzzer.
[1] https://webrtc-review.googlesource.com/c/src/+/51920
[2] https://webrtc-review.googlesource.com/c/src/+/52381
Bug: webrtc:7949
Change-Id: Idb10cc3ca9d6d2e4ac5824cc3391ed8aa680f6cd
Reviewed-on: https://webrtc-review.googlesource.com/54361
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Reviewed-by: Sam Zackrisson <saza@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#22103}
This CL adds the Level Estimator of the new gain controller. The Level
Estimator divides a 10ms input frame in kSubFramesInFrame=20 sub
frames. We take the maximal sample values in every sub frame. We then
apply attack/decay smoothing. This is the final level estimate.
The results will be used with InterpolatedGainCurve (see this CL
https://webrtc-review.googlesource.com/c/src/+/51920). For every level
estimate value, we look up a gain with
InterpolatedGainCurve::LookUpGainToApply. This gain is then applied to
the signal.
Bug: webrtc:7949
Change-Id: I2b4b3894a3e945d3dd916ce516c79abacb2b18b1
Reviewed-on: https://webrtc-review.googlesource.com/52381
Reviewed-by: Sam Zackrisson <saza@webrtc.org>
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#22054}
The FixedGainController (FGC) applies a fixed gain. It will also
control the limiter. The limiter will be landed over the next several
CLs.
The GainController2 is a 'private submodule' of APM. It will control
the new automatic gain controller (AGC). It controls the AGC through
Initialize() and ApplyConfig().
This CL contains
* build changes to make modules/audio_processing/agc2 an independent
target
* a new MutableFloatAudioFrame which is the audio interface between
AGC2 and APM
* move of the fixed gain application from GainController2 to
FixedGainController.
If you are a googler, there is more information in this doc:
https://docs.google.com/document/d/1RV2Doet3MZtUPAHVva61Vjo20iyd1bmmm3aR8znWpzo/edit#
Bug: webrtc:7949
Change-Id: Ief95cbbce83c3aafe54638fd2ab881c9fb8bdc3a
Reviewed-on: https://webrtc-review.googlesource.com/50440
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Reviewed-by: Oskar Sundbom <ossu@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#22046}
