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}
diff --git a/modules/audio_processing/audio_processing_impl.cc b/modules/audio_processing/audio_processing_impl.cc
index d28a44b..e92ae6d 100644
--- a/modules/audio_processing/audio_processing_impl.cc
+++ b/modules/audio_processing/audio_processing_impl.cc
@@ -69,29 +69,6 @@
"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.
int SuitableProcessRate(int minimum_rate,
int max_splitting_rate,
@@ -325,17 +302,44 @@
return error_code;
}
-const absl::optional<AudioProcessingImpl::GainController2ConfigOverride>
-GetGainController2ConfigOverride() {
+using DownmixMethod = AudioProcessing::Config::Pipeline::DownmixMethod;
+
+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";
if (!field_trial::IsEnabled(kFieldTrialName)) {
return absl::nullopt;
}
- constexpr InputVolumeController::Config kDefaultInputVolumeControllerConfig;
-
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(
"clipped_level_min",
kDefaultInputVolumeControllerConfig.clipped_level_min, 0, 255);
@@ -369,9 +373,9 @@
"speech_ratio_threshold",
kDefaultInputVolumeControllerConfig.speech_ratio_threshold, 0, 1);
+ // AGC2 adaptive digital controller configuration.
constexpr AudioProcessing::Config::GainController2::AdaptiveDigital
kDefaultAdaptiveDigitalConfig;
-
FieldTrialConstrained<double> headroom_db(
"headroom_db", kDefaultAdaptiveDigitalConfig.headroom_db, 0,
absl::nullopt);
@@ -390,83 +394,102 @@
kDefaultAdaptiveDigitalConfig.max_output_noise_level_dbfs, absl::nullopt,
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
// digital configs.
- const std::string field_trial_name =
- field_trial::FindFullName(kFieldTrialName);
-
ParseFieldTrial(
- {&enabled, &clipped_level_min, &clipped_level_step,
- &clipped_ratio_threshold, &clipped_wait_frames,
+ {&enabled, &switch_to_agc2, &min_input_volume, &clipped_level_min,
+ &clipped_level_step, &clipped_ratio_threshold, &clipped_wait_frames,
&enable_clipping_predictor, &target_range_max_dbfs,
&target_range_min_dbfs, &update_input_volume_wait_frames,
&speech_probability_threshold, &speech_ratio_threshold, &headroom_db,
&max_gain_db, &initial_gain_db, &max_gain_change_db_per_second,
- &max_output_noise_level_dbfs},
- field_trial_name);
-
+ &max_output_noise_level_dbfs, &disallow_transient_suppressor_usage},
+ field_trial::FindFullName(kFieldTrialName));
// Checked already by `IsEnabled()` before parsing, therefore always true.
RTC_DCHECK(enabled);
- return AudioProcessingImpl::GainController2ConfigOverride{
- .input_volume_controller_config =
- {
- .clipped_level_min = static_cast<int>(clipped_level_min.Get()),
- .clipped_level_step = static_cast<int>(clipped_level_step.Get()),
- .clipped_ratio_threshold =
- static_cast<float>(clipped_ratio_threshold.Get()),
- .clipped_wait_frames =
- static_cast<int>(clipped_wait_frames.Get()),
- .enable_clipping_predictor =
- static_cast<bool>(enable_clipping_predictor.Get()),
- .target_range_max_dbfs =
- static_cast<int>(target_range_max_dbfs.Get()),
- .target_range_min_dbfs =
- static_cast<int>(target_range_min_dbfs.Get()),
- .update_input_volume_wait_frames =
- static_cast<int>(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_config =
- {
- .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()),
- },
- };
+ const bool do_not_change_agc_config = !switch_to_agc2.Get();
+ if (do_not_change_agc_config && !disallow_transient_suppressor_usage.Get()) {
+ // Return an unspecifed value since, in this case, both the AGC2 and TS
+ // configurations won't be adjusted.
+ return absl::nullopt;
+ }
+ using Params = AudioProcessingImpl::GainController2ExperimentParams;
+ if (do_not_change_agc_config) {
+ // Return a value that leaves the AGC2 config unchanged and that always
+ // disables TS.
+ return Params{.agc2_config = absl::nullopt,
+ .disallow_transient_suppressor_usage = true};
+ }
+ // Return a value that switches all the gain control to AGC2.
+ return Params{
+ .agc2_config =
+ Params::Agc2Config{
+ .input_volume_controller =
+ {
+ .min_input_volume = min_input_volume.Get(),
+ .clipped_level_min = clipped_level_min.Get(),
+ .clipped_level_step = clipped_level_step.Get(),
+ .clipped_ratio_threshold =
+ static_cast<float>(clipped_ratio_threshold.Get()),
+ .clipped_wait_frames = clipped_wait_frames.Get(),
+ .enable_clipping_predictor =
+ enable_clipping_predictor.Get(),
+ .target_range_max_dbfs = target_range_max_dbfs.Get(),
+ .target_range_min_dbfs = target_range_min_dbfs.Get(),
+ .update_input_volume_wait_frames =
+ 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
-// suppression. When `gain_controller2_config_override` is specified,
-// switches all gain control to AGC2.
-AudioProcessing::Config AdjustConfig(
+AudioProcessing::Config AudioProcessingImpl::AdjustConfig(
const AudioProcessing::Config& config,
- bool disallow_transient_supporessor_usage,
- const absl::optional<AudioProcessingImpl::GainController2ConfigOverride>&
- gain_controller2_config_override) {
+ const absl::optional<AudioProcessingImpl::GainController2ExperimentParams>&
+ experiment_params) {
+ 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;
// Override the transient suppressor configuration.
- if (disallow_transient_supporessor_usage) {
+ if (experiment_params->disallow_transient_suppressor_usage) {
adjusted_config.transient_suppression.enabled = false;
}
// Override the auto gain control configuration if the AGC1 analog gain
- // controller is active and `gain_controller2_config_override` is
- // specified.
+ // controller is active and `experiment_params->agc2_config` is specified.
const bool agc1_analog_enabled =
config.gain_controller1.enabled &&
(config.gain_controller1.mode ==
AudioProcessing::Config::GainController1::kAdaptiveAnalog ||
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.
const bool hybrid_agc_config_detected =
config.gain_controller1.enabled &&
@@ -499,7 +522,7 @@
adjusted_config.gain_controller2.enabled = true;
adjusted_config.gain_controller2.input_volume_controller.enabled = true;
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;
}
}
@@ -507,26 +530,21 @@
return adjusted_config;
}
-using DownmixMethod = AudioProcessing::Config::Pipeline::DownmixMethod;
-
-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;
+TransientSuppressor::VadMode AudioProcessingImpl::GetTransientSuppressorVadMode(
+ const absl::optional<AudioProcessingImpl::GainController2ExperimentParams>&
+ params) {
+ if (params.has_value() && params->agc2_config.has_value() &&
+ !params->disallow_transient_suppressor_usage) {
+ // When the experiment is active, the gain control switches to AGC2 and TS
+ // can be active, use the RNN VAD to control TS. This choice will also
+ // disable the internal RNN VAD in AGC2.
+ return TransientSuppressor::VadMode::kRnnVad;
}
+ // 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(
bool capture_post_processor_enabled,
bool render_pre_processor_enabled,
@@ -644,20 +662,17 @@
: data_dumper_(new ApmDataDumper(instance_count_.fetch_add(1) + 1)),
use_setup_specific_default_aec3_config_(
UseSetupSpecificDefaultAec3Congfig()),
- gain_controller2_config_override_(GetGainController2ConfigOverride()),
+ gain_controller2_experiment_params_(GetGainController2ExperimentParams()),
use_denormal_disabler_(
!field_trial::IsEnabled("WebRTC-ApmDenormalDisablerKillSwitch")),
- disallow_transient_supporessor_usage_(
- field_trial::IsEnabled("WebRTC-ApmTransientSuppressorKillSwitch")),
- transient_suppressor_vad_mode_(GetTransientSuppressorVadMode()),
+ transient_suppressor_vad_mode_(
+ GetTransientSuppressorVadMode(gain_controller2_experiment_params_)),
capture_runtime_settings_(RuntimeSettingQueueSize()),
render_runtime_settings_(RuntimeSettingQueueSize()),
capture_runtime_settings_enqueuer_(&capture_runtime_settings_),
render_runtime_settings_enqueuer_(&render_runtime_settings_),
echo_control_factory_(std::move(echo_control_factory)),
- config_(AdjustConfig(config,
- disallow_transient_supporessor_usage_,
- gain_controller2_config_override_)),
+ config_(AdjustConfig(config, gain_controller2_experiment_params_)),
submodule_states_(!!capture_post_processor,
!!render_pre_processor,
!!capture_analyzer),
@@ -893,8 +908,7 @@
MutexLock lock_capture(&mutex_capture_);
const auto adjusted_config =
- AdjustConfig(config, disallow_transient_supporessor_usage_,
- gain_controller2_config_override_);
+ AdjustConfig(config, gain_controller2_experiment_params_);
RTC_LOG(LS_INFO) << "AudioProcessing::ApplyConfig: "
<< adjusted_config.ToString();
@@ -2340,11 +2354,16 @@
if (!submodules_.gain_controller2 || config_has_changed) {
const bool use_internal_vad =
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>(
- config_.gain_controller2,
- gain_controller2_config_override_.has_value()
- ? gain_controller2_config_override_->input_volume_controller_config
- : InputVolumeController::Config{},
+ config_.gain_controller2, input_volume_controller_config,
proc_fullband_sample_rate_hz(), num_proc_channels(), use_internal_vad);
submodules_.gain_controller2->SetCaptureOutputUsed(
capture_.capture_output_used);