Unify AGC2 experiment field trials into one

In order to experiment with AGC2 and TS at the same time, 3 field
trials are removed and merged into `WebRTC-Audio-GainController2`,
which is existing.

New parameters for the `WebRTC-Audio-GainController2` field trial:
- `switch_to_agc2`: true by default; when true, the gain control
  switches to AGC2 (both for the input volume and for the adaptive
  digital gain);
- `min_input_volume`: minimum input volume enforced by the input
  volume controller when the applied input volume is not zero;
- `disallow_transient_suppressor_usage`: when true, TS is never
  created.

Removed field trials:
- `WebRTC-Audio-Agc2-MinInputVolume`: now a parameter of
  `WebRTC-Audio-GainController2`;
- `WebRTC-ApmTransientSuppressorKillSwitch`: now a parameter of
  `WebRTC-Audio-GainController2`;
- `WebRTC-Audio-TransientSuppressorVadMode`: automatically inferred
  from `WebRTC-Audio-GainController2`.

Bug: webrtc:7494
Change-Id: I452798c0862d71f9adae6d163fe841df05ca44d6
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/287861
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Hanna Silen <silen@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#38890}
This commit is contained in:
Alessio Bazzica 2022-12-14 16:36:10 +01:00 committed by WebRTC LUCI CQ
parent 597a2ba41a
commit 3b51cd328e
6 changed files with 795 additions and 1189 deletions

View File

@ -50,29 +50,6 @@ Agc1ClippingPredictorConfig CreateClippingPredictorConfig(bool enabled) {
return config; return config;
} }
// Returns the minimum input volume to recommend.
// If the "WebRTC-Audio-Agc2-MinInputVolume" field trial is specified, parses it
// and returns the value specified after "Enabled-" if valid - i.e., in the
// range 0-255. Otherwise returns the default value.
// Example:
// "WebRTC-Audio-Agc2-MinInputVolume/Enabled-80" => returns 80.
int GetMinInputVolume() {
constexpr int kDefaultMinInputVolume = 12;
constexpr char kFieldTrial[] = "WebRTC-Audio-Agc2-MinInputVolume";
if (!webrtc::field_trial::IsEnabled(kFieldTrial)) {
return kDefaultMinInputVolume;
}
std::string field_trial_str = webrtc::field_trial::FindFullName(kFieldTrial);
int min_input_volume = -1;
sscanf(field_trial_str.c_str(), "Enabled-%d", &min_input_volume);
if (min_input_volume >= 0 && min_input_volume <= 255) {
return min_input_volume;
}
RTC_LOG(LS_WARNING) << "[AGC2] Invalid volume for " << kFieldTrial
<< ", ignored.";
return kDefaultMinInputVolume;
}
// Returns an input volume in the [`min_input_volume`, `kMaxInputVolume`] range // Returns an input volume in the [`min_input_volume`, `kMaxInputVolume`] range
// that reduces `gain_error_db`, which is a gain error estimated when // that reduces `gain_error_db`, which is a gain error estimated when
// `input_volume` was applied, according to a fixed gain map. // `input_volume` was applied, according to a fixed gain map.
@ -377,7 +354,7 @@ void MonoInputVolumeController::UpdateInputVolume(int rms_error_db) {
InputVolumeController::InputVolumeController(int num_capture_channels, InputVolumeController::InputVolumeController(int num_capture_channels,
const Config& config) const Config& config)
: num_capture_channels_(num_capture_channels), : num_capture_channels_(num_capture_channels),
min_input_volume_(GetMinInputVolume()), min_input_volume_(config.min_input_volume),
capture_output_used_(true), capture_output_used_(true),
clipped_level_step_(config.clipped_level_step), clipped_level_step_(config.clipped_level_step),
clipped_ratio_threshold_(config.clipped_ratio_threshold), clipped_ratio_threshold_(config.clipped_ratio_threshold),

View File

@ -35,6 +35,9 @@ class InputVolumeController final {
public: public:
// Config for the constructor. // Config for the constructor.
struct Config { struct Config {
// Minimum input volume that can be recommended. Not enforced when the
// applied input volume is zero outside startup.
int min_input_volume = 20;
// Lowest input volume level that will be applied in response to clipping. // Lowest input volume level that will be applied in response to clipping.
int clipped_level_min = 70; int clipped_level_min = 70;
// Amount input volume level is lowered with every clipping event. Limited // Amount input volume level is lowered with every clipping event. Limited
@ -52,13 +55,9 @@ class InputVolumeController final {
// [`target_range_min_dbfs`, `target_range_max_dbfs`], no input volume // [`target_range_min_dbfs`, `target_range_max_dbfs`], no input volume
// adjustments are done based on the speech level. For speech levels below // adjustments are done based on the speech level. For speech levels below
// and above the range, the targets `target_range_min_dbfs` and // and above the range, the targets `target_range_min_dbfs` and
// `target_range_max_dbfs` are used, respectively. The example values // `target_range_max_dbfs` are used, respectively.
// `target_range_max_dbfs` -18 and `target_range_min_dbfs` -48 refer to a int target_range_max_dbfs = -30;
// configuration where the zero-digital-gain target is -18 dBFS and the int target_range_min_dbfs = -50;
// digital gain control is expected to compensate for speech level errors
// up to -30 dB.
int target_range_max_dbfs = -18;
int target_range_min_dbfs = -48;
// Number of wait frames between the recommended input volume updates. // Number of wait frames between the recommended input volume updates.
int update_input_volume_wait_frames = 100; int update_input_volume_wait_frames = 100;
// Speech probability threshold: speech probabilities below the threshold // Speech probability threshold: speech probabilities below the threshold

View File

@ -38,7 +38,7 @@ constexpr int kNumChannels = 1;
constexpr int kInitialInputVolume = 128; constexpr int kInitialInputVolume = 128;
constexpr int kClippedMin = 165; // Arbitrary, but different from the default. constexpr int kClippedMin = 165; // Arbitrary, but different from the default.
constexpr float kAboveClippedThreshold = 0.2f; constexpr float kAboveClippedThreshold = 0.2f;
constexpr int kMinMicLevel = 12; constexpr int kMinMicLevel = 20;
constexpr int kClippedLevelStep = 15; constexpr int kClippedLevelStep = 15;
constexpr float kClippedRatioThreshold = 0.1f; constexpr float kClippedRatioThreshold = 0.1f;
constexpr int kClippedWaitFrames = 300; constexpr int kClippedWaitFrames = 300;
@ -56,7 +56,6 @@ using ClippingPredictorConfig = AudioProcessing::Config::GainController1::
using InputVolumeControllerConfig = InputVolumeController::Config; using InputVolumeControllerConfig = InputVolumeController::Config;
constexpr InputVolumeControllerConfig kDefaultInputVolumeControllerConfig{};
constexpr ClippingPredictorConfig kDefaultClippingPredictorConfig{}; constexpr ClippingPredictorConfig kDefaultClippingPredictorConfig{};
std::unique_ptr<InputVolumeController> CreateInputVolumeController( std::unique_ptr<InputVolumeController> CreateInputVolumeController(
@ -66,6 +65,7 @@ std::unique_ptr<InputVolumeController> CreateInputVolumeController(
bool enable_clipping_predictor = false, bool enable_clipping_predictor = false,
int update_input_volume_wait_frames = 0) { int update_input_volume_wait_frames = 0) {
InputVolumeControllerConfig config{ InputVolumeControllerConfig config{
.min_input_volume = kMinMicLevel,
.clipped_level_min = kClippedMin, .clipped_level_min = kClippedMin,
.clipped_level_step = clipped_level_step, .clipped_level_step = clipped_level_step,
.clipped_ratio_threshold = clipped_ratio_threshold, .clipped_ratio_threshold = clipped_ratio_threshold,
@ -82,34 +82,6 @@ std::unique_ptr<InputVolumeController> CreateInputVolumeController(
config); config);
} }
constexpr char kMinInputVolumeFieldTrial[] = "WebRTC-Audio-Agc2-MinInputVolume";
std::string GetAgcMinInputVolumeFieldTrial(const std::string& value) {
char field_trial_buffer[64];
rtc::SimpleStringBuilder builder(field_trial_buffer);
builder << kMinInputVolumeFieldTrial << "/" << value << "/";
return builder.str();
}
std::string GetAgcMinInputVolumeFieldTrialEnabled(
int enabled_value,
const std::string& suffix = "") {
RTC_DCHECK_GE(enabled_value, 0);
RTC_DCHECK_LE(enabled_value, 255);
char field_trial_buffer[64];
rtc::SimpleStringBuilder builder(field_trial_buffer);
builder << kMinInputVolumeFieldTrial << "/Enabled-" << enabled_value << suffix
<< "/";
return builder.str();
}
std::string GetAgcMinInputVolumeFieldTrial(absl::optional<int> volume) {
if (volume.has_value()) {
return GetAgcMinInputVolumeFieldTrialEnabled(*volume);
}
return GetAgcMinInputVolumeFieldTrial("Disabled");
}
// (Over)writes `samples_value` for the samples in `audio_buffer`. // (Over)writes `samples_value` for the samples in `audio_buffer`.
// When `clipped_ratio`, a value in [0, 1], is greater than 0, the corresponding // When `clipped_ratio`, a value in [0, 1], is greater than 0, the corresponding
// fraction of the frame is set to a full scale value to simulate clipping. // fraction of the frame is set to a full scale value to simulate clipping.
@ -150,31 +122,6 @@ void WriteAlternatingAudioBufferSamples(float samples_value,
} }
} }
// Deprecated.
// TODO(bugs.webrtc.org/7494): Delete this helper, use
// `InputVolumeControllerTestHelper::CallAgcSequence()` instead.
int CallAnalyzeAndRecommend(int num_calls,
int initial_volume,
const AudioBuffer& audio_buffer,
float speech_probability,
absl::optional<float> speech_level_dbfs,
InputVolumeController& controller) {
RTC_DCHECK(controller.capture_output_used());
int volume = initial_volume;
for (int n = 0; n < num_calls; ++n) {
controller.AnalyzeInputAudio(volume, audio_buffer);
const auto recommended_input_volume =
controller.RecommendInputVolume(speech_probability, speech_level_dbfs);
// Expect no errors: Applied volume set for every frame;
// `RecommendInputVolume()` returns a non-empty value.
EXPECT_TRUE(recommended_input_volume.has_value());
volume = *recommended_input_volume;
}
return volume;
}
// Reads a given number of 10 ms chunks from a PCM file and feeds them to // Reads a given number of 10 ms chunks from a PCM file and feeds them to
// `InputVolumeController`. // `InputVolumeController`.
class SpeechSamplesReader { class SpeechSamplesReader {
@ -379,24 +326,12 @@ class InputVolumeControllerTestHelper {
}; };
class InputVolumeControllerParametrizedTest class InputVolumeControllerParametrizedTest
: public ::testing::TestWithParam<absl::optional<int>> { : public ::testing::TestWithParam<int> {};
protected:
InputVolumeControllerParametrizedTest()
: field_trials_(GetAgcMinInputVolumeFieldTrial(GetParam())) {}
int GetMinInputVolume() const { return GetParam().value_or(kMinMicLevel); }
private:
test::ScopedFieldTrials field_trials_;
};
INSTANTIATE_TEST_SUITE_P(,
InputVolumeControllerParametrizedTest,
::testing::Values(absl::nullopt, 12, 20));
TEST_P(InputVolumeControllerParametrizedTest, TEST_P(InputVolumeControllerParametrizedTest,
StartupMinVolumeConfigurationIsRespected) { StartupMinVolumeConfigurationIsRespected) {
InputVolumeControllerTestHelper helper; InputVolumeControllerTestHelper helper(
/*config=*/{.min_input_volume = GetParam()});
EXPECT_EQ(*helper.CallAgcSequence(kInitialInputVolume, kHighSpeechProbability, EXPECT_EQ(*helper.CallAgcSequence(kInitialInputVolume, kHighSpeechProbability,
kSpeechLevel), kSpeechLevel),
@ -404,7 +339,9 @@ TEST_P(InputVolumeControllerParametrizedTest,
} }
TEST_P(InputVolumeControllerParametrizedTest, MicVolumeResponseToRmsError) { TEST_P(InputVolumeControllerParametrizedTest, MicVolumeResponseToRmsError) {
InputVolumeControllerTestHelper helper; InputVolumeControllerConfig config = GetInputVolumeControllerTestConfig();
config.min_input_volume = GetParam();
InputVolumeControllerTestHelper helper(config);
int volume = *helper.CallAgcSequence(kInitialInputVolume, int volume = *helper.CallAgcSequence(kInitialInputVolume,
kHighSpeechProbability, kSpeechLevel); kHighSpeechProbability, kSpeechLevel);
@ -446,7 +383,10 @@ TEST_P(InputVolumeControllerParametrizedTest, MicVolumeResponseToRmsError) {
} }
TEST_P(InputVolumeControllerParametrizedTest, MicVolumeIsLimited) { TEST_P(InputVolumeControllerParametrizedTest, MicVolumeIsLimited) {
InputVolumeControllerTestHelper helper; InputVolumeControllerConfig config = GetInputVolumeControllerTestConfig();
const int min_input_volume = GetParam();
config.min_input_volume = min_input_volume;
InputVolumeControllerTestHelper helper(config);
int volume = *helper.CallAgcSequence(kInitialInputVolume, int volume = *helper.CallAgcSequence(kInitialInputVolume,
kHighSpeechProbability, kSpeechLevel); kHighSpeechProbability, kSpeechLevel);
@ -492,16 +432,18 @@ TEST_P(InputVolumeControllerParametrizedTest, MicVolumeIsLimited) {
// Won't go lower than the minimum. // Won't go lower than the minimum.
volume = helper.CallRecommendInputVolume(/*num_calls=*/1, volume, volume = helper.CallRecommendInputVolume(/*num_calls=*/1, volume,
kHighSpeechProbability, 22.0f); kHighSpeechProbability, 22.0f);
EXPECT_EQ(volume, std::max(18, GetMinInputVolume())); EXPECT_EQ(volume, std::max(18, min_input_volume));
volume = helper.CallRecommendInputVolume(/*num_calls=*/1, volume, volume = helper.CallRecommendInputVolume(/*num_calls=*/1, volume,
kHighSpeechProbability, 22.0f); kHighSpeechProbability, 22.0f);
EXPECT_EQ(volume, std::max(12, GetMinInputVolume())); EXPECT_EQ(volume, std::max(12, min_input_volume));
} }
TEST_P(InputVolumeControllerParametrizedTest, NoActionWhileMuted) { TEST_P(InputVolumeControllerParametrizedTest, NoActionWhileMuted) {
InputVolumeControllerTestHelper helper_1; InputVolumeControllerTestHelper helper_1(
InputVolumeControllerTestHelper helper_2; /*config=*/{.min_input_volume = GetParam()});
InputVolumeControllerTestHelper helper_2(
/*config=*/{.min_input_volume = GetParam()});
int volume_1 = *helper_1.CallAgcSequence(/*applied_input_volume=*/255, int volume_1 = *helper_1.CallAgcSequence(/*applied_input_volume=*/255,
kHighSpeechProbability, kSpeechLevel, kHighSpeechProbability, kSpeechLevel,
@ -531,7 +473,8 @@ TEST_P(InputVolumeControllerParametrizedTest, NoActionWhileMuted) {
TEST_P(InputVolumeControllerParametrizedTest, TEST_P(InputVolumeControllerParametrizedTest,
UnmutingChecksVolumeWithoutRaising) { UnmutingChecksVolumeWithoutRaising) {
InputVolumeControllerTestHelper helper; InputVolumeControllerTestHelper helper(
/*config=*/{.min_input_volume = GetParam()});
helper.CallAgcSequence(kInitialInputVolume, kHighSpeechProbability, helper.CallAgcSequence(kInitialInputVolume, kHighSpeechProbability,
kSpeechLevel); kSpeechLevel);
@ -548,7 +491,9 @@ TEST_P(InputVolumeControllerParametrizedTest,
} }
TEST_P(InputVolumeControllerParametrizedTest, UnmutingRaisesTooLowVolume) { TEST_P(InputVolumeControllerParametrizedTest, UnmutingRaisesTooLowVolume) {
InputVolumeControllerTestHelper helper; const int min_input_volume = GetParam();
InputVolumeControllerTestHelper helper(
/*config=*/{.min_input_volume = min_input_volume});
helper.CallAgcSequence(kInitialInputVolume, kHighSpeechProbability, helper.CallAgcSequence(kInitialInputVolume, kHighSpeechProbability,
kSpeechLevel); kSpeechLevel);
@ -560,12 +505,14 @@ TEST_P(InputVolumeControllerParametrizedTest, UnmutingRaisesTooLowVolume) {
EXPECT_EQ( EXPECT_EQ(
helper.CallRecommendInputVolume(/*num_calls=*/1, kInputVolume, helper.CallRecommendInputVolume(/*num_calls=*/1, kInputVolume,
kHighSpeechProbability, kSpeechLevel), kHighSpeechProbability, kSpeechLevel),
GetMinInputVolume()); min_input_volume);
} }
TEST_P(InputVolumeControllerParametrizedTest, TEST_P(InputVolumeControllerParametrizedTest,
ManualLevelChangeResultsInNoSetMicCall) { ManualLevelChangeResultsInNoSetMicCall) {
InputVolumeControllerTestHelper helper; InputVolumeControllerConfig config = GetInputVolumeControllerTestConfig();
config.min_input_volume = GetParam();
InputVolumeControllerTestHelper helper(config);
int volume = *helper.CallAgcSequence(kInitialInputVolume, int volume = *helper.CallAgcSequence(kInitialInputVolume,
kHighSpeechProbability, kSpeechLevel); kHighSpeechProbability, kSpeechLevel);
@ -589,7 +536,9 @@ TEST_P(InputVolumeControllerParametrizedTest,
TEST_P(InputVolumeControllerParametrizedTest, TEST_P(InputVolumeControllerParametrizedTest,
RecoveryAfterManualLevelChangeFromMax) { RecoveryAfterManualLevelChangeFromMax) {
InputVolumeControllerTestHelper helper; InputVolumeControllerConfig config = GetInputVolumeControllerTestConfig();
config.min_input_volume = GetParam();
InputVolumeControllerTestHelper helper(config);
int volume = *helper.CallAgcSequence(kInitialInputVolume, int volume = *helper.CallAgcSequence(kInitialInputVolume,
kHighSpeechProbability, kSpeechLevel); kHighSpeechProbability, kSpeechLevel);
@ -621,7 +570,10 @@ TEST_P(InputVolumeControllerParametrizedTest,
// of the input volume. // of the input volume.
TEST_P(InputVolumeControllerParametrizedTest, TEST_P(InputVolumeControllerParametrizedTest,
EnforceMinInputVolumeDuringUpwardsAdjustment) { EnforceMinInputVolumeDuringUpwardsAdjustment) {
InputVolumeControllerTestHelper helper; const int min_input_volume = GetParam();
InputVolumeControllerConfig config = GetInputVolumeControllerTestConfig();
config.min_input_volume = min_input_volume;
InputVolumeControllerTestHelper helper(config);
int volume = *helper.CallAgcSequence(kInitialInputVolume, int volume = *helper.CallAgcSequence(kInitialInputVolume,
kHighSpeechProbability, kSpeechLevel); kHighSpeechProbability, kSpeechLevel);
@ -631,19 +583,19 @@ TEST_P(InputVolumeControllerParametrizedTest,
/*num_calls=*/1, /*initial_volume=*/1, kHighSpeechProbability, -17.0f); /*num_calls=*/1, /*initial_volume=*/1, kHighSpeechProbability, -17.0f);
// Trigger an upward adjustment of the input volume. // Trigger an upward adjustment of the input volume.
EXPECT_EQ(volume, GetMinInputVolume()); EXPECT_EQ(volume, min_input_volume);
volume = helper.CallRecommendInputVolume(/*num_calls=*/1, volume, volume = helper.CallRecommendInputVolume(/*num_calls=*/1, volume,
kHighSpeechProbability, -29.0f); kHighSpeechProbability, -29.0f);
EXPECT_EQ(volume, GetMinInputVolume()); EXPECT_EQ(volume, min_input_volume);
volume = helper.CallRecommendInputVolume(/*num_calls=*/1, volume, volume = helper.CallRecommendInputVolume(/*num_calls=*/1, volume,
kHighSpeechProbability, -30.0f); kHighSpeechProbability, -30.0f);
EXPECT_EQ(volume, GetMinInputVolume()); EXPECT_EQ(volume, min_input_volume);
// After a number of consistently low speech level observations, the input // After a number of consistently low speech level observations, the input
// volume is eventually raised above the minimum. // volume is eventually raised above the minimum.
volume = helper.CallRecommendInputVolume(/*num_calls=*/10, volume, volume = helper.CallRecommendInputVolume(/*num_calls=*/10, volume,
kHighSpeechProbability, -38.0f); kHighSpeechProbability, -38.0f);
EXPECT_GT(volume, GetMinInputVolume()); EXPECT_GT(volume, min_input_volume);
} }
// Checks that, when the min mic level override is specified, AGC immediately // Checks that, when the min mic level override is specified, AGC immediately
@ -651,7 +603,9 @@ TEST_P(InputVolumeControllerParametrizedTest,
// minimum gain to enforce. // minimum gain to enforce.
TEST_P(InputVolumeControllerParametrizedTest, TEST_P(InputVolumeControllerParametrizedTest,
RecoveryAfterManualLevelChangeBelowMin) { RecoveryAfterManualLevelChangeBelowMin) {
InputVolumeControllerTestHelper helper; const int min_input_volume = GetParam();
InputVolumeControllerTestHelper helper(
/*config=*/{.min_input_volume = min_input_volume});
int volume = *helper.CallAgcSequence(kInitialInputVolume, int volume = *helper.CallAgcSequence(kInitialInputVolume,
kHighSpeechProbability, kSpeechLevel); kHighSpeechProbability, kSpeechLevel);
@ -659,11 +613,12 @@ TEST_P(InputVolumeControllerParametrizedTest,
// AGC won't take any action. // AGC won't take any action.
volume = helper.CallRecommendInputVolume( volume = helper.CallRecommendInputVolume(
/*num_calls=*/1, /*initial_volume=*/1, kHighSpeechProbability, -17.0f); /*num_calls=*/1, /*initial_volume=*/1, kHighSpeechProbability, -17.0f);
EXPECT_EQ(volume, GetMinInputVolume()); EXPECT_EQ(volume, min_input_volume);
} }
TEST_P(InputVolumeControllerParametrizedTest, NoClippingHasNoImpact) { TEST_P(InputVolumeControllerParametrizedTest, NoClippingHasNoImpact) {
InputVolumeControllerTestHelper helper; InputVolumeControllerTestHelper helper(
/*config=*/{.min_input_volume = GetParam()});
helper.CallAgcSequence(kInitialInputVolume, kHighSpeechProbability, helper.CallAgcSequence(kInitialInputVolume, kHighSpeechProbability,
kSpeechLevel); kSpeechLevel);
@ -673,7 +628,8 @@ TEST_P(InputVolumeControllerParametrizedTest, NoClippingHasNoImpact) {
TEST_P(InputVolumeControllerParametrizedTest, TEST_P(InputVolumeControllerParametrizedTest,
ClippingUnderThresholdHasNoImpact) { ClippingUnderThresholdHasNoImpact) {
InputVolumeControllerTestHelper helper; InputVolumeControllerTestHelper helper(
/*config=*/{.min_input_volume = GetParam()});
helper.CallAgcSequence(kInitialInputVolume, kHighSpeechProbability, helper.CallAgcSequence(kInitialInputVolume, kHighSpeechProbability,
kSpeechLevel); kSpeechLevel);
@ -682,7 +638,8 @@ TEST_P(InputVolumeControllerParametrizedTest,
} }
TEST_P(InputVolumeControllerParametrizedTest, ClippingLowersVolume) { TEST_P(InputVolumeControllerParametrizedTest, ClippingLowersVolume) {
InputVolumeControllerTestHelper helper; InputVolumeControllerTestHelper helper(
/*config=*/{.min_input_volume = GetParam()});
helper.CallAgcSequence(/*applied_input_volume=*/255, kHighSpeechProbability, helper.CallAgcSequence(/*applied_input_volume=*/255, kHighSpeechProbability,
kSpeechLevel); kSpeechLevel);
@ -692,7 +649,8 @@ TEST_P(InputVolumeControllerParametrizedTest, ClippingLowersVolume) {
TEST_P(InputVolumeControllerParametrizedTest, TEST_P(InputVolumeControllerParametrizedTest,
WaitingPeriodBetweenClippingChecks) { WaitingPeriodBetweenClippingChecks) {
InputVolumeControllerTestHelper helper; InputVolumeControllerTestHelper helper(
/*config=*/{.min_input_volume = GetParam()});
helper.CallAgcSequence(/*applied_input_volume=*/255, kHighSpeechProbability, helper.CallAgcSequence(/*applied_input_volume=*/255, kHighSpeechProbability,
kSpeechLevel); kSpeechLevel);
@ -710,7 +668,9 @@ TEST_P(InputVolumeControllerParametrizedTest,
} }
TEST_P(InputVolumeControllerParametrizedTest, ClippingLoweringIsLimited) { TEST_P(InputVolumeControllerParametrizedTest, ClippingLoweringIsLimited) {
InputVolumeControllerTestHelper helper; InputVolumeControllerConfig config = GetInputVolumeControllerTestConfig();
config.min_input_volume = GetParam();
InputVolumeControllerTestHelper helper(config);
helper.CallAgcSequence(/*applied_input_volume=*/180, kHighSpeechProbability, helper.CallAgcSequence(/*applied_input_volume=*/180, kHighSpeechProbability,
kSpeechLevel); kSpeechLevel);
@ -725,7 +685,8 @@ TEST_P(InputVolumeControllerParametrizedTest, ClippingLoweringIsLimited) {
TEST_P(InputVolumeControllerParametrizedTest, TEST_P(InputVolumeControllerParametrizedTest,
ClippingMaxIsRespectedWhenEqualToLevel) { ClippingMaxIsRespectedWhenEqualToLevel) {
InputVolumeControllerTestHelper helper; InputVolumeControllerTestHelper helper(
/*config=*/{.min_input_volume = GetParam()});
helper.CallAgcSequence(/*applied_input_volume=*/255, kHighSpeechProbability, helper.CallAgcSequence(/*applied_input_volume=*/255, kHighSpeechProbability,
kSpeechLevel); kSpeechLevel);
@ -740,7 +701,9 @@ TEST_P(InputVolumeControllerParametrizedTest,
TEST_P(InputVolumeControllerParametrizedTest, TEST_P(InputVolumeControllerParametrizedTest,
ClippingMaxIsRespectedWhenHigherThanLevel) { ClippingMaxIsRespectedWhenHigherThanLevel) {
InputVolumeControllerTestHelper helper; InputVolumeControllerConfig config = GetInputVolumeControllerTestConfig();
config.min_input_volume = GetParam();
InputVolumeControllerTestHelper helper(config);
helper.CallAgcSequence(/*applied_input_volume=*/200, kHighSpeechProbability, helper.CallAgcSequence(/*applied_input_volume=*/200, kHighSpeechProbability,
kSpeechLevel); kSpeechLevel);
@ -758,7 +721,9 @@ TEST_P(InputVolumeControllerParametrizedTest,
} }
TEST_P(InputVolumeControllerParametrizedTest, UserCanRaiseVolumeAfterClipping) { TEST_P(InputVolumeControllerParametrizedTest, UserCanRaiseVolumeAfterClipping) {
InputVolumeControllerTestHelper helper; InputVolumeControllerConfig config = GetInputVolumeControllerTestConfig();
config.min_input_volume = GetParam();
InputVolumeControllerTestHelper helper(config);
helper.CallAgcSequence(/*applied_input_volume=*/225, kHighSpeechProbability, helper.CallAgcSequence(/*applied_input_volume=*/225, kHighSpeechProbability,
kSpeechLevel); kSpeechLevel);
@ -787,7 +752,9 @@ TEST_P(InputVolumeControllerParametrizedTest, UserCanRaiseVolumeAfterClipping) {
TEST_P(InputVolumeControllerParametrizedTest, TEST_P(InputVolumeControllerParametrizedTest,
ClippingDoesNotPullLowVolumeBackUp) { ClippingDoesNotPullLowVolumeBackUp) {
InputVolumeControllerTestHelper helper; InputVolumeControllerConfig config = GetInputVolumeControllerTestConfig();
config.min_input_volume = GetParam();
InputVolumeControllerTestHelper helper(config);
helper.CallAgcSequence(/*applied_input_volume=*/80, kHighSpeechProbability, helper.CallAgcSequence(/*applied_input_volume=*/80, kHighSpeechProbability,
kSpeechLevel); kSpeechLevel);
@ -798,7 +765,8 @@ TEST_P(InputVolumeControllerParametrizedTest,
} }
TEST_P(InputVolumeControllerParametrizedTest, TakesNoActionOnZeroMicVolume) { TEST_P(InputVolumeControllerParametrizedTest, TakesNoActionOnZeroMicVolume) {
InputVolumeControllerTestHelper helper; InputVolumeControllerTestHelper helper(
/*config=*/{.min_input_volume = GetParam()});
helper.CallAgcSequence(kInitialInputVolume, kHighSpeechProbability, helper.CallAgcSequence(kInitialInputVolume, kHighSpeechProbability,
kSpeechLevel); kSpeechLevel);
@ -809,7 +777,9 @@ TEST_P(InputVolumeControllerParametrizedTest, TakesNoActionOnZeroMicVolume) {
} }
TEST_P(InputVolumeControllerParametrizedTest, ClippingDetectionLowersVolume) { TEST_P(InputVolumeControllerParametrizedTest, ClippingDetectionLowersVolume) {
InputVolumeControllerTestHelper helper; InputVolumeControllerConfig config = GetInputVolumeControllerTestConfig();
config.min_input_volume = GetParam();
InputVolumeControllerTestHelper helper(config);
int volume = *helper.CallAgcSequence(/*applied_input_volume=*/255, int volume = *helper.CallAgcSequence(/*applied_input_volume=*/255,
kHighSpeechProbability, kSpeechLevel, kHighSpeechProbability, kSpeechLevel,
/*num_calls=*/1); /*num_calls=*/1);
@ -829,298 +799,6 @@ TEST_P(InputVolumeControllerParametrizedTest, ClippingDetectionLowersVolume) {
EXPECT_EQ(volume, 240); EXPECT_EQ(volume, 240);
} }
TEST(InputVolumeControllerTest, MinInputVolumeDefault) {
std::unique_ptr<InputVolumeController> controller =
CreateInputVolumeController(kClippedLevelStep, kClippedRatioThreshold,
kClippedWaitFrames);
EXPECT_EQ(controller->channel_controllers_[0]->min_input_volume(),
kMinMicLevel);
}
TEST(InputVolumeControllerTest, MinInputVolumeDisabled) {
for (const std::string& field_trial_suffix : {"", "_20220210"}) {
test::ScopedFieldTrials field_trial(
GetAgcMinInputVolumeFieldTrial("Disabled" + field_trial_suffix));
std::unique_ptr<InputVolumeController> controller =
CreateInputVolumeController(kClippedLevelStep, kClippedRatioThreshold,
kClippedWaitFrames);
EXPECT_EQ(controller->channel_controllers_[0]->min_input_volume(),
kMinMicLevel);
}
}
// Checks that a field-trial parameter outside of the valid range [0,255] is
// ignored.
TEST(InputVolumeControllerTest, MinInputVolumeOutOfRangeAbove) {
test::ScopedFieldTrials field_trial(
GetAgcMinInputVolumeFieldTrial("Enabled-256"));
std::unique_ptr<InputVolumeController> controller =
CreateInputVolumeController(kClippedLevelStep, kClippedRatioThreshold,
kClippedWaitFrames);
EXPECT_EQ(controller->channel_controllers_[0]->min_input_volume(),
kMinMicLevel);
}
// Checks that a field-trial parameter outside of the valid range [0,255] is
// ignored.
TEST(InputVolumeControllerTest, MinInputVolumeOutOfRangeBelow) {
test::ScopedFieldTrials field_trial(
GetAgcMinInputVolumeFieldTrial("Enabled--1"));
std::unique_ptr<InputVolumeController> controller =
CreateInputVolumeController(kClippedLevelStep, kClippedRatioThreshold,
kClippedWaitFrames);
EXPECT_EQ(controller->channel_controllers_[0]->min_input_volume(),
kMinMicLevel);
}
// Verifies that a valid experiment changes the minimum microphone level. The
// start volume is larger than the min level and should therefore not be
// changed.
TEST(InputVolumeControllerTest, MinInputVolumeEnabled50) {
constexpr int kMinInputVolume = 50;
for (const std::string& field_trial_suffix : {"", "_20220210"}) {
SCOPED_TRACE(field_trial_suffix);
test::ScopedFieldTrials field_trial(GetAgcMinInputVolumeFieldTrialEnabled(
kMinInputVolume, field_trial_suffix));
std::unique_ptr<InputVolumeController> controller =
CreateInputVolumeController(kClippedLevelStep, kClippedRatioThreshold,
kClippedWaitFrames);
EXPECT_EQ(controller->channel_controllers_[0]->min_input_volume(),
kMinInputVolume);
}
}
// Checks that, when the "WebRTC-Audio-Agc2-MinInputVolume" field trial is
// specified with a valid value, the mic level never gets lowered beyond the
// override value in the presence of clipping.
TEST(InputVolumeControllerTest, MinInputVolumeCheckMinLevelWithClipping) {
constexpr int kMinInputVolume = 250;
// Create and initialize two AGCs by specifying and leaving unspecified the
// relevant field trial.
const auto factory = []() {
std::unique_ptr<InputVolumeController> controller =
CreateInputVolumeController(kClippedLevelStep, kClippedRatioThreshold,
kClippedWaitFrames);
controller->Initialize();
return controller;
};
std::unique_ptr<InputVolumeController> controller = factory();
std::unique_ptr<InputVolumeController> controller_with_override;
{
test::ScopedFieldTrials field_trial(
GetAgcMinInputVolumeFieldTrialEnabled(kMinInputVolume));
controller_with_override = factory();
}
// Create a test input signal which containts 80% of clipped samples.
AudioBuffer audio_buffer(kSampleRateHz, 1, kSampleRateHz, 1, kSampleRateHz,
1);
WriteAudioBufferSamples(/*samples_value=*/4000.0f, /*clipped_ratio=*/0.8f,
audio_buffer);
// Simulate 4 seconds of clipping; it is expected to trigger a downward
// adjustment of the analog gain. Use low speech probability to limit the
// volume changes to clipping handling.
const int volume = CallAnalyzeAndRecommend(
/*num_calls=*/400, kInitialInputVolume, audio_buffer,
kLowSpeechProbability, /*speech_level_dbfs=*/-42.0f, *controller);
const int volume_with_override = CallAnalyzeAndRecommend(
/*num_calls=*/400, kInitialInputVolume, audio_buffer,
kLowSpeechProbability, /*speech_level_dbfs=*/-42.0f,
*controller_with_override);
// Make sure that an adaptation occurred.
ASSERT_GT(volume, 0);
// Check that the test signal triggers a larger downward adaptation for
// `controller`, which is allowed to reach a lower gain.
EXPECT_GT(volume_with_override, volume);
// Check that the gain selected by `controller_with_override` equals the
// minimum value overridden via field trial.
EXPECT_EQ(volume_with_override, kMinInputVolume);
}
// Checks that, when the "WebRTC-Audio-Agc2-MinInputVolume" field trial is
// specified with a valid value, the mic level never gets lowered beyond the
// override value in the presence of clipping when RMS error is not empty.
// TODO(webrtc:7494): Revisit the test after moving the number of update wait
// frames to APM config. The test passes but internally the gain update timing
// differs.
TEST(InputVolumeControllerTest,
MinInputVolumeCheckMinLevelWithClippingWithRmsError) {
constexpr int kMinInputVolume = 250;
// Create and initialize two AGCs by specifying and leaving unspecified the
// relevant field trial.
const auto factory = []() {
std::unique_ptr<InputVolumeController> controller =
CreateInputVolumeController(kClippedLevelStep, kClippedRatioThreshold,
kClippedWaitFrames);
controller->Initialize();
return controller;
};
std::unique_ptr<InputVolumeController> controller = factory();
std::unique_ptr<InputVolumeController> controller_with_override;
{
test::ScopedFieldTrials field_trial(
GetAgcMinInputVolumeFieldTrialEnabled(kMinInputVolume));
controller_with_override = factory();
}
// Create a test input signal which containts 80% of clipped samples.
AudioBuffer audio_buffer(kSampleRateHz, 1, kSampleRateHz, 1, kSampleRateHz,
1);
WriteAudioBufferSamples(/*samples_value=*/4000.0f, /*clipped_ratio=*/0.8f,
audio_buffer);
// Simulate 4 seconds of clipping; it is expected to trigger a downward
// adjustment of the analog gain.
const int volume = CallAnalyzeAndRecommend(
/*num_calls=*/400, kInitialInputVolume, audio_buffer,
kHighSpeechProbability,
/*speech_level_dbfs=*/-18.0f, *controller);
const int volume_with_override = CallAnalyzeAndRecommend(
/*num_calls=*/400, kInitialInputVolume, audio_buffer,
kHighSpeechProbability,
/*speech_level_dbfs=*/-18.0f, *controller_with_override);
// Make sure that an adaptation occurred.
ASSERT_GT(volume, 0);
// Check that the test signal triggers a larger downward adaptation for
// `controller`, which is allowed to reach a lower gain.
EXPECT_GT(volume_with_override, volume);
// Check that the gain selected by `controller_with_override` equals the
// minimum value overridden via field trial.
EXPECT_EQ(volume_with_override, kMinInputVolume);
}
// Checks that, when the "WebRTC-Audio-Agc2-MinInputVolume" field trial is
// specified with a value lower than the `clipped_level_min`, the behavior of
// the analog gain controller is the same as that obtained when the field trial
// is not specified.
TEST(InputVolumeControllerTest, MinInputVolumeCompareMicLevelWithClipping) {
// Create and initialize two AGCs by specifying and leaving unspecified the
// relevant field trial.
const auto factory = []() {
// Use a large clipped level step to more quickly decrease the analog gain
// with clipping.
InputVolumeControllerConfig config = kDefaultInputVolumeControllerConfig;
config.clipped_level_step = 64;
config.clipped_ratio_threshold = kClippedRatioThreshold;
config.clipped_wait_frames = kClippedWaitFrames;
auto controller = std::make_unique<InputVolumeController>(
/*num_capture_channels=*/1, config);
controller->Initialize();
return controller;
};
std::unique_ptr<InputVolumeController> controller = factory();
std::unique_ptr<InputVolumeController> controller_with_override;
{
constexpr int kMinInputVolume = 20;
static_assert(kDefaultInputVolumeControllerConfig.clipped_level_min >=
kMinInputVolume,
"Use a lower override value.");
test::ScopedFieldTrials field_trial(
GetAgcMinInputVolumeFieldTrialEnabled(kMinInputVolume));
controller_with_override = factory();
}
// Create a test input signal which containts 80% of clipped samples.
AudioBuffer audio_buffer(kSampleRateHz, 1, kSampleRateHz, 1, kSampleRateHz,
1);
WriteAudioBufferSamples(/*samples_value=*/4000.0f, /*clipped_ratio=*/0.8f,
audio_buffer);
// Simulate 4 seconds of clipping; it is expected to trigger a downward
// adjustment of the analog gain. Use low speech probability to limit the
// volume changes to clipping handling.
const int volume = CallAnalyzeAndRecommend(
/*num_calls=*/400, kInitialInputVolume, audio_buffer,
kLowSpeechProbability, /*speech_level_dbfs=*/-18, *controller);
const int volume_with_override = CallAnalyzeAndRecommend(
/*num_calls=*/400, kInitialInputVolume, audio_buffer,
kLowSpeechProbability, /*speech_level_dbfs=*/-18,
*controller_with_override);
// Make sure that an adaptation occurred.
ASSERT_GT(volume, 0);
// Check that the selected analog gain is the same for both controllers and
// that it equals the minimum level reached when clipping is handled. That is
// expected because the minimum microphone level override is less than the
// minimum level used when clipping is detected.
EXPECT_EQ(volume, volume_with_override);
EXPECT_EQ(volume_with_override,
kDefaultInputVolumeControllerConfig.clipped_level_min);
}
// Checks that, when the "WebRTC-Audio-Agc2-MinInputVolume" field trial is
// specified with a value lower than the `clipped_level_min`, the behavior of
// the analog gain controller is the same as that obtained when the field trial
// is not specified.
// TODO(webrtc:7494): Revisit the test after moving the number of update wait
// frames to APM config. The test passes but internally the gain update timing
// differs.
TEST(InputVolumeControllerTest,
MinInputVolumeCompareMicLevelWithClippingWithRmsError) {
// Create and initialize two AGCs by specifying and leaving unspecified the
// relevant field trial.
const auto factory = []() {
// Use a large clipped level step to more quickly decrease the analog gain
// with clipping.
InputVolumeControllerConfig config = kDefaultInputVolumeControllerConfig;
config.clipped_level_step = 64;
config.clipped_ratio_threshold = kClippedRatioThreshold;
config.clipped_wait_frames = kClippedWaitFrames;
auto controller = std::make_unique<InputVolumeController>(
/*num_capture_channels=*/1, config);
controller->Initialize();
return controller;
};
std::unique_ptr<InputVolumeController> controller = factory();
std::unique_ptr<InputVolumeController> controller_with_override;
{
constexpr int kMinInputVolume = 20;
static_assert(kDefaultInputVolumeControllerConfig.clipped_level_min >=
kMinInputVolume,
"Use a lower override value.");
test::ScopedFieldTrials field_trial(
GetAgcMinInputVolumeFieldTrialEnabled(kMinInputVolume));
controller_with_override = factory();
}
// Create a test input signal which containts 80% of clipped samples.
AudioBuffer audio_buffer(kSampleRateHz, 1, kSampleRateHz, 1, kSampleRateHz,
1);
WriteAudioBufferSamples(/*samples_value=*/4000.0f, /*clipped_ratio=*/0.8f,
audio_buffer);
const int volume = CallAnalyzeAndRecommend(
/*num_calls=*/400, kInitialInputVolume, audio_buffer,
/*speech_probability=*/0.7f,
/*speech_level_dbfs=*/-18.0f, *controller);
const int volume_with_override = CallAnalyzeAndRecommend(
/*num_calls=*/400, kInitialInputVolume, audio_buffer,
/*speech_probability=*/0.7f,
/*speech_level_dbfs=*/-18.0f, *controller_with_override);
// Make sure that an adaptation occurred.
ASSERT_GT(volume, 0);
// Check that the selected analog gain is the same for both controllers and
// that it equals the minimum level reached when clipping is handled. That is
// expected because the minimum microphone level override is less than the
// minimum level used when clipping is detected.
EXPECT_EQ(volume, volume_with_override);
EXPECT_EQ(volume_with_override,
kDefaultInputVolumeControllerConfig.clipped_level_min);
}
// TODO(bugs.webrtc.org/12774): Test the bahavior of `clipped_level_step`. // TODO(bugs.webrtc.org/12774): Test the bahavior of `clipped_level_step`.
// TODO(bugs.webrtc.org/12774): Test the bahavior of `clipped_ratio_threshold`. // TODO(bugs.webrtc.org/12774): Test the bahavior of `clipped_ratio_threshold`.
// TODO(bugs.webrtc.org/12774): Test the bahavior of `clipped_wait_frames`. // TODO(bugs.webrtc.org/12774): Test the bahavior of `clipped_wait_frames`.
@ -1416,6 +1094,42 @@ TEST(InputVolumeControllerTest, UpdateInputVolumeWaitFramesIsEffective) {
ASSERT_GT(volume_wait_100, kInputVolume); ASSERT_GT(volume_wait_100, kInputVolume);
} }
INSTANTIATE_TEST_SUITE_P(,
InputVolumeControllerParametrizedTest,
::testing::Values(12, 20));
TEST(InputVolumeControllerTest,
MinInputVolumeEnforcedWithClippingWhenAboveClippedLevelMin) {
InputVolumeControllerTestHelper helper(
/*config=*/{.min_input_volume = 80, .clipped_level_min = 70});
// Trigger a downward adjustment caused by clipping input. Use a low speech
// probability to limit the volume changes to clipping handling.
WriteAudioBufferSamples(/*samples_value=*/4000.0f, /*clipped_ratio=*/0.8f,
helper.audio_buffer);
constexpr int kNumCalls = 800;
helper.CallAgcSequence(/*applied_input_volume=*/100, kLowSpeechProbability,
/*speech_level_dbfs=*/-18.0f, kNumCalls);
EXPECT_EQ(helper.controller.recommended_input_volume(), 80);
}
TEST(InputVolumeControllerTest,
ClippedlevelMinEnforcedWithClippingWhenAboveMinInputVolume) {
InputVolumeControllerTestHelper helper(
/*config=*/{.min_input_volume = 70, .clipped_level_min = 80});
// Trigger a downward adjustment caused by clipping input. Use a low speech
// probability to limit the volume changes to clipping handling.
WriteAudioBufferSamples(/*samples_value=*/4000.0f, /*clipped_ratio=*/0.8f,
helper.audio_buffer);
constexpr int kNumCalls = 800;
helper.CallAgcSequence(/*applied_input_volume=*/100, kLowSpeechProbability,
/*speech_level_dbfs=*/-18.0f, kNumCalls);
EXPECT_EQ(helper.controller.recommended_input_volume(), 80);
}
TEST(InputVolumeControllerTest, SpeechRatioThresholdIsEffective) { TEST(InputVolumeControllerTest, SpeechRatioThresholdIsEffective) {
constexpr int kInputVolume = kInitialInputVolume; constexpr int kInputVolume = kInitialInputVolume;
// Create two input volume controllers with 10 frames between volume updates // Create two input volume controllers with 10 frames between volume updates

View File

@ -69,29 +69,6 @@ bool UseSetupSpecificDefaultAec3Congfig() {
"WebRTC-Aec3SetupSpecificDefaultConfigDefaultsKillSwitch"); "WebRTC-Aec3SetupSpecificDefaultConfigDefaultsKillSwitch");
} }
// If the "WebRTC-Audio-TransientSuppressorVadMode" field trial is unspecified,
// returns `TransientSuppressor::VadMode::kDefault`, otherwise parses the field
// trial and returns the specified mode:
// - WebRTC-Audio-TransientSuppressorVadMode/Enabled-Default returns `kDefault`;
// - WebRTC-Audio-TransientSuppressorVadMode/Enabled-RnnVad returns `kRnnVad`;
// - WebRTC-Audio-TransientSuppressorVadMode/Enabled-NoVad returns `kNoVad`.
TransientSuppressor::VadMode GetTransientSuppressorVadMode() {
constexpr char kFieldTrial[] = "WebRTC-Audio-TransientSuppressorVadMode";
std::string full_name = webrtc::field_trial::FindFullName(kFieldTrial);
if (full_name.empty() || absl::EndsWith(full_name, "-Default")) {
return TransientSuppressor::VadMode::kDefault;
}
if (absl::EndsWith(full_name, "-RnnVad")) {
return TransientSuppressor::VadMode::kRnnVad;
}
if (absl::EndsWith(full_name, "-NoVad")) {
return TransientSuppressor::VadMode::kNoVad;
}
// Fallback to default.
RTC_LOG(LS_WARNING) << "Invalid parameter for " << kFieldTrial;
return TransientSuppressor::VadMode::kDefault;
}
// Identify the native processing rate that best handles a sample rate. // Identify the native processing rate that best handles a sample rate.
int SuitableProcessRate(int minimum_rate, int SuitableProcessRate(int minimum_rate,
int max_splitting_rate, int max_splitting_rate,
@ -325,17 +302,44 @@ int HandleUnsupportedAudioFormats(const float* const* src,
return error_code; return error_code;
} }
const absl::optional<AudioProcessingImpl::GainController2ConfigOverride> using DownmixMethod = AudioProcessing::Config::Pipeline::DownmixMethod;
GetGainController2ConfigOverride() {
void SetDownmixMethod(AudioBuffer& buffer, DownmixMethod method) {
switch (method) {
case DownmixMethod::kAverageChannels:
buffer.set_downmixing_by_averaging();
break;
case DownmixMethod::kUseFirstChannel:
buffer.set_downmixing_to_specific_channel(/*channel=*/0);
break;
}
}
constexpr int kUnspecifiedDataDumpInputVolume = -100;
} // namespace
// Throughout webrtc, it's assumed that success is represented by zero.
static_assert(AudioProcessing::kNoError == 0, "kNoError must be zero");
absl::optional<AudioProcessingImpl::GainController2ExperimentParams>
AudioProcessingImpl::GetGainController2ExperimentParams() {
constexpr char kFieldTrialName[] = "WebRTC-Audio-GainController2"; constexpr char kFieldTrialName[] = "WebRTC-Audio-GainController2";
if (!field_trial::IsEnabled(kFieldTrialName)) { if (!field_trial::IsEnabled(kFieldTrialName)) {
return absl::nullopt; return absl::nullopt;
} }
constexpr InputVolumeController::Config kDefaultInputVolumeControllerConfig;
FieldTrialFlag enabled("Enabled", false); FieldTrialFlag enabled("Enabled", false);
// Whether the gain control should switch to AGC2. Enabled by default.
FieldTrialParameter<bool> switch_to_agc2("switch_to_agc2", true);
// AGC2 input volume controller configuration.
constexpr InputVolumeController::Config kDefaultInputVolumeControllerConfig;
FieldTrialConstrained<int> min_input_volume(
"min_input_volume", kDefaultInputVolumeControllerConfig.min_input_volume,
0, 255);
FieldTrialConstrained<int> clipped_level_min( FieldTrialConstrained<int> clipped_level_min(
"clipped_level_min", "clipped_level_min",
kDefaultInputVolumeControllerConfig.clipped_level_min, 0, 255); kDefaultInputVolumeControllerConfig.clipped_level_min, 0, 255);
@ -369,9 +373,9 @@ GetGainController2ConfigOverride() {
"speech_ratio_threshold", "speech_ratio_threshold",
kDefaultInputVolumeControllerConfig.speech_ratio_threshold, 0, 1); kDefaultInputVolumeControllerConfig.speech_ratio_threshold, 0, 1);
// AGC2 adaptive digital controller configuration.
constexpr AudioProcessing::Config::GainController2::AdaptiveDigital constexpr AudioProcessing::Config::GainController2::AdaptiveDigital
kDefaultAdaptiveDigitalConfig; kDefaultAdaptiveDigitalConfig;
FieldTrialConstrained<double> headroom_db( FieldTrialConstrained<double> headroom_db(
"headroom_db", kDefaultAdaptiveDigitalConfig.headroom_db, 0, "headroom_db", kDefaultAdaptiveDigitalConfig.headroom_db, 0,
absl::nullopt); absl::nullopt);
@ -390,83 +394,102 @@ GetGainController2ConfigOverride() {
kDefaultAdaptiveDigitalConfig.max_output_noise_level_dbfs, absl::nullopt, kDefaultAdaptiveDigitalConfig.max_output_noise_level_dbfs, absl::nullopt,
0); 0);
// Transient suppressor.
FieldTrialParameter<bool> disallow_transient_suppressor_usage(
"disallow_transient_suppressor_usage", false);
// Field-trial based override for the input volume controller and adaptive // Field-trial based override for the input volume controller and adaptive
// digital configs. // digital configs.
const std::string field_trial_name =
field_trial::FindFullName(kFieldTrialName);
ParseFieldTrial( ParseFieldTrial(
{&enabled, &clipped_level_min, &clipped_level_step, {&enabled, &switch_to_agc2, &min_input_volume, &clipped_level_min,
&clipped_ratio_threshold, &clipped_wait_frames, &clipped_level_step, &clipped_ratio_threshold, &clipped_wait_frames,
&enable_clipping_predictor, &target_range_max_dbfs, &enable_clipping_predictor, &target_range_max_dbfs,
&target_range_min_dbfs, &update_input_volume_wait_frames, &target_range_min_dbfs, &update_input_volume_wait_frames,
&speech_probability_threshold, &speech_ratio_threshold, &headroom_db, &speech_probability_threshold, &speech_ratio_threshold, &headroom_db,
&max_gain_db, &initial_gain_db, &max_gain_change_db_per_second, &max_gain_db, &initial_gain_db, &max_gain_change_db_per_second,
&max_output_noise_level_dbfs}, &max_output_noise_level_dbfs, &disallow_transient_suppressor_usage},
field_trial_name); field_trial::FindFullName(kFieldTrialName));
// Checked already by `IsEnabled()` before parsing, therefore always true. // Checked already by `IsEnabled()` before parsing, therefore always true.
RTC_DCHECK(enabled); RTC_DCHECK(enabled);
return AudioProcessingImpl::GainController2ConfigOverride{ const bool do_not_change_agc_config = !switch_to_agc2.Get();
.input_volume_controller_config = if (do_not_change_agc_config && !disallow_transient_suppressor_usage.Get()) {
{ // Return an unspecifed value since, in this case, both the AGC2 and TS
.clipped_level_min = static_cast<int>(clipped_level_min.Get()), // configurations won't be adjusted.
.clipped_level_step = static_cast<int>(clipped_level_step.Get()), return absl::nullopt;
.clipped_ratio_threshold = }
static_cast<float>(clipped_ratio_threshold.Get()), using Params = AudioProcessingImpl::GainController2ExperimentParams;
.clipped_wait_frames = if (do_not_change_agc_config) {
static_cast<int>(clipped_wait_frames.Get()), // Return a value that leaves the AGC2 config unchanged and that always
.enable_clipping_predictor = // disables TS.
static_cast<bool>(enable_clipping_predictor.Get()), return Params{.agc2_config = absl::nullopt,
.target_range_max_dbfs = .disallow_transient_suppressor_usage = true};
static_cast<int>(target_range_max_dbfs.Get()), }
.target_range_min_dbfs = // Return a value that switches all the gain control to AGC2.
static_cast<int>(target_range_min_dbfs.Get()), return Params{
.update_input_volume_wait_frames = .agc2_config =
static_cast<int>(update_input_volume_wait_frames.Get()), Params::Agc2Config{
.speech_probability_threshold = .input_volume_controller =
static_cast<float>(speech_probability_threshold.Get()), {
.speech_ratio_threshold = .min_input_volume = min_input_volume.Get(),
static_cast<float>(speech_ratio_threshold.Get()), .clipped_level_min = clipped_level_min.Get(),
}, .clipped_level_step = clipped_level_step.Get(),
.adaptive_digital_config = .clipped_ratio_threshold =
{ static_cast<float>(clipped_ratio_threshold.Get()),
.headroom_db = static_cast<float>(headroom_db.Get()), .clipped_wait_frames = clipped_wait_frames.Get(),
.max_gain_db = static_cast<float>(max_gain_db.Get()), .enable_clipping_predictor =
.initial_gain_db = static_cast<float>(initial_gain_db.Get()), enable_clipping_predictor.Get(),
.max_gain_change_db_per_second = .target_range_max_dbfs = target_range_max_dbfs.Get(),
static_cast<float>(max_gain_change_db_per_second.Get()), .target_range_min_dbfs = target_range_min_dbfs.Get(),
.max_output_noise_level_dbfs = .update_input_volume_wait_frames =
static_cast<float>(max_output_noise_level_dbfs.Get()), update_input_volume_wait_frames.Get(),
}, .speech_probability_threshold = static_cast<float>(
}; speech_probability_threshold.Get()),
.speech_ratio_threshold =
static_cast<float>(speech_ratio_threshold.Get()),
},
.adaptive_digital_controller =
{
.headroom_db = static_cast<float>(headroom_db.Get()),
.max_gain_db = static_cast<float>(max_gain_db.Get()),
.initial_gain_db =
static_cast<float>(initial_gain_db.Get()),
.max_gain_change_db_per_second = static_cast<float>(
max_gain_change_db_per_second.Get()),
.max_output_noise_level_dbfs =
static_cast<float>(max_output_noise_level_dbfs.Get()),
}},
.disallow_transient_suppressor_usage =
disallow_transient_suppressor_usage.Get()};
} }
// If `disallow_transient_supporessor_usage` is true, disables transient AudioProcessing::Config AudioProcessingImpl::AdjustConfig(
// suppression. When `gain_controller2_config_override` is specified,
// switches all gain control to AGC2.
AudioProcessing::Config AdjustConfig(
const AudioProcessing::Config& config, const AudioProcessing::Config& config,
bool disallow_transient_supporessor_usage, const absl::optional<AudioProcessingImpl::GainController2ExperimentParams>&
const absl::optional<AudioProcessingImpl::GainController2ConfigOverride>& experiment_params) {
gain_controller2_config_override) { if (!experiment_params.has_value() ||
(!experiment_params->agc2_config.has_value() &&
!experiment_params->disallow_transient_suppressor_usage)) {
// When the experiment parameters are unspecified or when the AGC and TS
// configuration are not overridden, return the unmodified configuration.
return config;
}
AudioProcessing::Config adjusted_config = config; AudioProcessing::Config adjusted_config = config;
// Override the transient suppressor configuration. // Override the transient suppressor configuration.
if (disallow_transient_supporessor_usage) { if (experiment_params->disallow_transient_suppressor_usage) {
adjusted_config.transient_suppression.enabled = false; adjusted_config.transient_suppression.enabled = false;
} }
// Override the auto gain control configuration if the AGC1 analog gain // Override the auto gain control configuration if the AGC1 analog gain
// controller is active and `gain_controller2_config_override` is // controller is active and `experiment_params->agc2_config` is specified.
// specified.
const bool agc1_analog_enabled = const bool agc1_analog_enabled =
config.gain_controller1.enabled && config.gain_controller1.enabled &&
(config.gain_controller1.mode == (config.gain_controller1.mode ==
AudioProcessing::Config::GainController1::kAdaptiveAnalog || AudioProcessing::Config::GainController1::kAdaptiveAnalog ||
config.gain_controller1.analog_gain_controller.enabled); config.gain_controller1.analog_gain_controller.enabled);
if (agc1_analog_enabled && gain_controller2_config_override.has_value()) { if (agc1_analog_enabled && experiment_params->agc2_config.has_value()) {
// Check that the unadjusted AGC config meets the preconditions. // Check that the unadjusted AGC config meets the preconditions.
const bool hybrid_agc_config_detected = const bool hybrid_agc_config_detected =
config.gain_controller1.enabled && config.gain_controller1.enabled &&
@ -499,7 +522,7 @@ AudioProcessing::Config AdjustConfig(
adjusted_config.gain_controller2.enabled = true; adjusted_config.gain_controller2.enabled = true;
adjusted_config.gain_controller2.input_volume_controller.enabled = true; adjusted_config.gain_controller2.input_volume_controller.enabled = true;
adjusted_config.gain_controller2.adaptive_digital = adjusted_config.gain_controller2.adaptive_digital =
gain_controller2_config_override->adaptive_digital_config; experiment_params->agc2_config->adaptive_digital_controller;
adjusted_config.gain_controller2.adaptive_digital.enabled = true; adjusted_config.gain_controller2.adaptive_digital.enabled = true;
} }
} }
@ -507,26 +530,21 @@ AudioProcessing::Config AdjustConfig(
return adjusted_config; return adjusted_config;
} }
using DownmixMethod = AudioProcessing::Config::Pipeline::DownmixMethod; TransientSuppressor::VadMode AudioProcessingImpl::GetTransientSuppressorVadMode(
const absl::optional<AudioProcessingImpl::GainController2ExperimentParams>&
void SetDownmixMethod(AudioBuffer& buffer, DownmixMethod method) { params) {
switch (method) { if (params.has_value() && params->agc2_config.has_value() &&
case DownmixMethod::kAverageChannels: !params->disallow_transient_suppressor_usage) {
buffer.set_downmixing_by_averaging(); // When the experiment is active, the gain control switches to AGC2 and TS
break; // can be active, use the RNN VAD to control TS. This choice will also
case DownmixMethod::kUseFirstChannel: // disable the internal RNN VAD in AGC2.
buffer.set_downmixing_to_specific_channel(/*channel=*/0); return TransientSuppressor::VadMode::kRnnVad;
break;
} }
// If TS is disabled, the returned value does not matter. If enabled, use the
// default VAD.
return TransientSuppressor::VadMode::kDefault;
} }
constexpr int kUnspecifiedDataDumpInputVolume = -100;
} // namespace
// Throughout webrtc, it's assumed that success is represented by zero.
static_assert(AudioProcessing::kNoError == 0, "kNoError must be zero");
AudioProcessingImpl::SubmoduleStates::SubmoduleStates( AudioProcessingImpl::SubmoduleStates::SubmoduleStates(
bool capture_post_processor_enabled, bool capture_post_processor_enabled,
bool render_pre_processor_enabled, bool render_pre_processor_enabled,
@ -644,20 +662,17 @@ AudioProcessingImpl::AudioProcessingImpl(
: data_dumper_(new ApmDataDumper(instance_count_.fetch_add(1) + 1)), : data_dumper_(new ApmDataDumper(instance_count_.fetch_add(1) + 1)),
use_setup_specific_default_aec3_config_( use_setup_specific_default_aec3_config_(
UseSetupSpecificDefaultAec3Congfig()), UseSetupSpecificDefaultAec3Congfig()),
gain_controller2_config_override_(GetGainController2ConfigOverride()), gain_controller2_experiment_params_(GetGainController2ExperimentParams()),
use_denormal_disabler_( use_denormal_disabler_(
!field_trial::IsEnabled("WebRTC-ApmDenormalDisablerKillSwitch")), !field_trial::IsEnabled("WebRTC-ApmDenormalDisablerKillSwitch")),
disallow_transient_supporessor_usage_( transient_suppressor_vad_mode_(
field_trial::IsEnabled("WebRTC-ApmTransientSuppressorKillSwitch")), GetTransientSuppressorVadMode(gain_controller2_experiment_params_)),
transient_suppressor_vad_mode_(GetTransientSuppressorVadMode()),
capture_runtime_settings_(RuntimeSettingQueueSize()), capture_runtime_settings_(RuntimeSettingQueueSize()),
render_runtime_settings_(RuntimeSettingQueueSize()), render_runtime_settings_(RuntimeSettingQueueSize()),
capture_runtime_settings_enqueuer_(&capture_runtime_settings_), capture_runtime_settings_enqueuer_(&capture_runtime_settings_),
render_runtime_settings_enqueuer_(&render_runtime_settings_), render_runtime_settings_enqueuer_(&render_runtime_settings_),
echo_control_factory_(std::move(echo_control_factory)), echo_control_factory_(std::move(echo_control_factory)),
config_(AdjustConfig(config, config_(AdjustConfig(config, gain_controller2_experiment_params_)),
disallow_transient_supporessor_usage_,
gain_controller2_config_override_)),
submodule_states_(!!capture_post_processor, submodule_states_(!!capture_post_processor,
!!render_pre_processor, !!render_pre_processor,
!!capture_analyzer), !!capture_analyzer),
@ -893,8 +908,7 @@ void AudioProcessingImpl::ApplyConfig(const AudioProcessing::Config& config) {
MutexLock lock_capture(&mutex_capture_); MutexLock lock_capture(&mutex_capture_);
const auto adjusted_config = const auto adjusted_config =
AdjustConfig(config, disallow_transient_supporessor_usage_, AdjustConfig(config, gain_controller2_experiment_params_);
gain_controller2_config_override_);
RTC_LOG(LS_INFO) << "AudioProcessing::ApplyConfig: " RTC_LOG(LS_INFO) << "AudioProcessing::ApplyConfig: "
<< adjusted_config.ToString(); << adjusted_config.ToString();
@ -2340,11 +2354,16 @@ void AudioProcessingImpl::InitializeGainController2(bool config_has_changed) {
if (!submodules_.gain_controller2 || config_has_changed) { if (!submodules_.gain_controller2 || config_has_changed) {
const bool use_internal_vad = const bool use_internal_vad =
transient_suppressor_vad_mode_ != TransientSuppressor::VadMode::kRnnVad; transient_suppressor_vad_mode_ != TransientSuppressor::VadMode::kRnnVad;
const bool input_volume_controller_config_overridden =
gain_controller2_experiment_params_.has_value() &&
gain_controller2_experiment_params_->agc2_config.has_value();
const InputVolumeController::Config input_volume_controller_config =
input_volume_controller_config_overridden
? gain_controller2_experiment_params_->agc2_config
->input_volume_controller
: InputVolumeController::Config{};
submodules_.gain_controller2 = std::make_unique<GainController2>( submodules_.gain_controller2 = std::make_unique<GainController2>(
config_.gain_controller2, config_.gain_controller2, input_volume_controller_config,
gain_controller2_config_override_.has_value()
? gain_controller2_config_override_->input_volume_controller_config
: InputVolumeController::Config{},
proc_fullband_sample_rate_hz(), num_proc_channels(), use_internal_vad); proc_fullband_sample_rate_hz(), num_proc_channels(), use_internal_vad);
submodules_.gain_controller2->SetCaptureOutputUsed( submodules_.gain_controller2->SetCaptureOutputUsed(
capture_.capture_output_used); capture_.capture_output_used);

View File

@ -138,14 +138,6 @@ class AudioProcessingImpl : public AudioProcessing {
AudioProcessing::Config GetConfig() const override; AudioProcessing::Config GetConfig() const override;
// TODO(bugs.webrtc.org/7494): Remove when the related field trial is
// removed.
struct GainController2ConfigOverride {
InputVolumeController::Config input_volume_controller_config;
AudioProcessing::Config::GainController2::AdaptiveDigital
adaptive_digital_config;
};
protected: protected:
// Overridden in a mock. // Overridden in a mock.
virtual void InitializeLocked() virtual void InitializeLocked()
@ -199,19 +191,47 @@ class AudioProcessingImpl : public AudioProcessing {
static std::atomic<int> instance_count_; static std::atomic<int> instance_count_;
const bool use_setup_specific_default_aec3_config_; const bool use_setup_specific_default_aec3_config_;
// TODO(bugs.webrtc.org/7494): Remove when the linked field trial is removed. // Parameters for the "GainController2" experiment which determines whether
// Override based on the "WebRTC-Audio-GainController2" field trial for the // the following APM sub-modules are created and, if so, their configurations:
// AGC2 input volume controller and adaptive digital controller configuration. // AGC2 (`gain_controller2`), AGC1 (`gain_control`, `agc_manager`) and TS
const absl::optional<GainController2ConfigOverride> // (`transient_suppressor`).
gain_controller2_config_override_; // TODO(bugs.webrtc.org/7494): Remove when the "WebRTC-Audio-GainController2"
// field trial is removed.
struct GainController2ExperimentParams {
struct Agc2Config {
InputVolumeController::Config input_volume_controller;
AudioProcessing::Config::GainController2::AdaptiveDigital
adaptive_digital_controller;
};
// When `agc2_config` is specified, all gain control switches to AGC2 and
// the configuration is overridden.
absl::optional<Agc2Config> agc2_config;
// When true, the transient suppressor submodule is never created regardless
// of the APM configuration.
bool disallow_transient_suppressor_usage;
};
// Specified when the "WebRTC-Audio-GainController2" field trial is specified.
// TODO(bugs.webrtc.org/7494): Remove when the "WebRTC-Audio-GainController2"
// field trial is removed.
const absl::optional<GainController2ExperimentParams>
gain_controller2_experiment_params_;
// Parses the "WebRTC-Audio-GainController2" field trial. If disabled, returns
// an unspecified value.
static absl::optional<GainController2ExperimentParams>
GetGainController2ExperimentParams();
// When `experiment_params` is specified, returns an APM configuration
// modified according to the experiment parameters. Otherwise returns
// `config`.
static AudioProcessing::Config AdjustConfig(
const AudioProcessing::Config& config,
const absl::optional<GainController2ExperimentParams>& experiment_params);
static TransientSuppressor::VadMode GetTransientSuppressorVadMode(
const absl::optional<GainController2ExperimentParams>& experiment_params);
const bool use_denormal_disabler_; const bool use_denormal_disabler_;
// When true, the transient suppressor submodule is never created regardless
// of the APM configuration.
// TODO(bugs.webrtc.org/13663): Remove when the linked field trial is removed.
const bool disallow_transient_supporessor_usage_;
const TransientSuppressor::VadMode transient_suppressor_vad_mode_; const TransientSuppressor::VadMode transient_suppressor_vad_mode_;
SwapQueue<RuntimeSetting> capture_runtime_settings_; SwapQueue<RuntimeSetting> capture_runtime_settings_;

File diff suppressed because it is too large Load Diff