webrtc_m130/modules/audio_processing/agc2/gain_curve_applier.cc
Alex Loiko 507e8d1f71 Reland of "Choose between APM-AGC-Limiter and Apm-AGC2-fixed-gain_controller."
The webrtc::AudioMixer uses a limiter component. This CL allows
changes the APM-AGC limiter to the APM-AGC2 limiter though a Chrome
field trial.

The AGC2 limiter has a float interface. We plan to eventually switch
to the AGC2 limiter. Therefore, we will now mix in de-interleaved
floats. Float mixing will happen both when using the old limiter and
when using the new one.

After this CL the mixer will support two limiters. The limiters have
different interfaces and need different processing steps. Because of
that, we make (rather big) changes to the control flow in
FrameCombiner. For a short while, we will mix in deinterleaved floats
when using any limiter.

Originally landed in https://webrtc-review.googlesource.com/c/src/+/56141/

Reverted in https://webrtc-review.googlesource.com/c/src/+/57940
because of both breaking compilation and having a severe error. The
error is fixed and a test is added. The compilation issue is fixed.

Bug: webrtc:8925
Change-Id: Ieba138dee9652c826459fe637ae2dccbbc06bcf0
Reviewed-on: https://webrtc-review.googlesource.com/58085
Reviewed-by: Gustaf Ullberg <gustaf@webrtc.org>
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#22207}
2018-02-27 15:47:39 +00:00

133 lines
5.0 KiB
C++

/*
* Copyright (c) 2018 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "modules/audio_processing/agc2/gain_curve_applier.h"
#include <algorithm>
#include <array>
#include <cmath>
#include "api/array_view.h"
#include "modules/audio_processing/logging/apm_data_dumper.h"
#include "rtc_base/checks.h"
namespace webrtc {
namespace {
// This constant affects the way scaling factors are interpolated for the first
// sub-frame of a frame. Only in the case in which the first sub-frame has an
// estimated level which is greater than the that of the previous analyzed
// sub-frame, linear interpolation is replaced with a power function which
// reduces the chances of over-shooting (and hence saturation), however reducing
// the fixed gain effectiveness.
constexpr float kAttackFirstSubframeInterpolationPower = 8.f;
void InterpolateFirstSubframe(float last_factor,
float current_factor,
rtc::ArrayView<float> subframe) {
const auto n = subframe.size();
constexpr auto p = kAttackFirstSubframeInterpolationPower;
for (size_t i = 0; i < n; ++i) {
subframe[i] = std::pow(1.f - i / n, p) * (last_factor - current_factor) +
current_factor;
}
}
void ComputePerSampleSubframeFactors(
const std::array<float, kSubFramesInFrame + 1>& scaling_factors,
size_t samples_per_channel,
rtc::ArrayView<float> per_sample_scaling_factors) {
const size_t num_subframes = scaling_factors.size() - 1;
const size_t subframe_size =
rtc::CheckedDivExact(samples_per_channel, num_subframes);
// Handle first sub-frame differently in case of attack.
const bool is_attack = scaling_factors[0] > scaling_factors[1];
if (is_attack) {
InterpolateFirstSubframe(
scaling_factors[0], scaling_factors[1],
rtc::ArrayView<float>(
per_sample_scaling_factors.subview(0, subframe_size)));
}
for (size_t i = is_attack ? 1 : 0; i < num_subframes; ++i) {
const size_t subframe_start = i * subframe_size;
const float scaling_start = scaling_factors[i];
const float scaling_end = scaling_factors[i + 1];
const float scaling_diff = (scaling_end - scaling_start) / subframe_size;
for (size_t j = 0; j < subframe_size; ++j) {
per_sample_scaling_factors[subframe_start + j] =
scaling_start + scaling_diff * j;
}
}
}
void ScaleSamples(rtc::ArrayView<const float> per_sample_scaling_factors,
AudioFrameView<float> signal) {
const size_t samples_per_channel = signal.samples_per_channel();
RTC_DCHECK_EQ(samples_per_channel, per_sample_scaling_factors.size());
for (size_t i = 0; i < signal.num_channels(); ++i) {
auto channel = signal.channel(i);
for (size_t j = 0; j < samples_per_channel; ++j) {
channel[j] *= per_sample_scaling_factors[j];
}
}
}
} // namespace
GainCurveApplier::GainCurveApplier(size_t sample_rate_hz,
ApmDataDumper* apm_data_dumper)
: interp_gain_curve_(apm_data_dumper),
level_estimator_(sample_rate_hz, apm_data_dumper),
apm_data_dumper_(apm_data_dumper) {}
GainCurveApplier::~GainCurveApplier() = default;
void GainCurveApplier::Process(AudioFrameView<float> signal) {
const auto level_estimate = level_estimator_.ComputeLevel(signal);
RTC_DCHECK_EQ(level_estimate.size() + 1, scaling_factors_.size());
scaling_factors_[0] = last_scaling_factor_;
std::transform(level_estimate.begin(), level_estimate.end(),
scaling_factors_.begin() + 1, [this](float x) {
return interp_gain_curve_.LookUpGainToApply(x);
});
const size_t samples_per_channel = signal.samples_per_channel();
RTC_DCHECK_LE(samples_per_channel, kMaximalNumberOfSamplesPerChannel);
auto per_sample_scaling_factors = rtc::ArrayView<float>(
&per_sample_scaling_factors_[0], samples_per_channel);
ComputePerSampleSubframeFactors(scaling_factors_, samples_per_channel,
per_sample_scaling_factors);
ScaleSamples(per_sample_scaling_factors, signal);
last_scaling_factor_ = scaling_factors_.back();
// Dump data for debug.
apm_data_dumper_->DumpRaw("agc2_gain_curve_applier_scaling_factors",
samples_per_channel,
per_sample_scaling_factors_.data());
}
InterpolatedGainCurve::Stats GainCurveApplier::GetGainCurveStats() const {
return interp_gain_curve_.get_stats();
}
void GainCurveApplier::SetSampleRate(size_t sample_rate_hz) {
level_estimator_.SetSampleRate(sample_rate_hz);
// Check that per_sample_scaling_factors_ is large enough.
RTC_DCHECK_LE(sample_rate_hz,
kMaximalNumberOfSamplesPerChannel * 1000 / kFrameDurationMs);
}
} // namespace webrtc