tg2sip/libtgvoip/webrtc_dsp/modules/audio_processing/agc2/limiter.cc

151 lines
5.5 KiB
C++
Raw Normal View History

Squashed 'libtgvoip/' changes from 6053cf5..cfd62e6 cfd62e6 Why did it change the OS X project 3a58a16 2.4.3 c4a48b3 Updated OS X project 564eada Fix #63 4f64e2e fixes 0c732e2 fixes 12e76ed better logging f015b79 Merge pull request #62 from xvitaly/big-endian a1df90f Set preferred audio session parameters on iOS 59a975b Fixes 8fd89fc Fixes, mic level testing and volume adjustment 243acfa Backported WebRTC upstream patch with Big Endian support. fed3bb7 Detect when proxy does not support UDP and persist that across calls a7546d4 Merge commit '6d03dd9ae4bf48d7344341cdd2d055ebd3a6a42e' into public 6d03dd9 version 69adf70 Use server config for APM + iOS crash fix 0b42ec8 Update iOS project f1b9e63 packet logging beeea45 I apparently still suck at C++ memory management 24fceba Update project 7f54b91 crash fix f85ce99 Save more data in data saving mode f4c4f79 Collect packet stats and accept json string for server config 78e584c New protocol version: optimized packet size 8cf9177 Fixed build on iOS 9dd089d fixed build on android 5caaaaf Updated WebRTC APM cc0cf35 fixed deadlock 02f4835 Rearranged VoIPController methods and added sections 912f73d Updated OS X project 39376df Fixed audio glitches on Windows dfe1f03 Updated project 81daf3f fix 296187a Merge pull request #58 from telegramdesktop/tdesktop 44956ac Merge pull request #57 from UnigramDev/public fb0a2b0 Fix build for Linux. d6cf1b7 Updated UWP wrapper 0f06289 Merge branch 'public' of github.com:grishka/libtgvoip into public dcfad91 Fix #54 162f447 Merge pull request #56 from telegramdesktop/tdesktop a7ee511 Merge remote-tracking branch 'origin/tdesktop' into HEAD 467b148 Removed unused files b1a0b3d 2.3 9b292fd Fix warning in Xcode 10. 8d8522a Merge pull request #53 from UnigramDev/public 646f7d6 Merge branch 'public' into public 14d782b Fixes 68acf59 Added GetSignalBarsCount and GetConnectionState to CXWrapper 761c586 Added GetStats to CXWrapper f643b02 Prevent crash if UWP WASAPI devices aren't found b2ac10e Fixed UWP project 9a1ec51 Fixed build for Windows Phone, fixed some warnings 4aea54f fix git-subtree-dir: libtgvoip git-subtree-split: cfd62e66a825348ac51f49e5d20bf8827fef7a38
2019-02-06 18:22:38 +00:00
/*
* 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/limiter.h"
#include <algorithm>
#include <array>
#include <cmath>
#include "api/array_view.h"
#include "modules/audio_processing/agc2/agc2_common.h"
#include "modules/audio_processing/logging/apm_data_dumper.h"
#include "rtc_base/checks.h"
#include "rtc_base/numerics/safe_minmax.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] = rtc::SafeClamp(channel[j] * per_sample_scaling_factors[j],
kMinFloatS16Value, kMaxFloatS16Value);
}
}
}
void CheckLimiterSampleRate(size_t sample_rate_hz) {
// Check that per_sample_scaling_factors_ is large enough.
RTC_DCHECK_LE(sample_rate_hz,
kMaximalNumberOfSamplesPerChannel * 1000 / kFrameDurationMs);
}
} // namespace
Limiter::Limiter(size_t sample_rate_hz,
ApmDataDumper* apm_data_dumper,
std::string histogram_name)
: interp_gain_curve_(apm_data_dumper, histogram_name),
level_estimator_(sample_rate_hz, apm_data_dumper),
apm_data_dumper_(apm_data_dumper) {
CheckLimiterSampleRate(sample_rate_hz);
}
Limiter::~Limiter() = default;
void Limiter::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 Limiter::GetGainCurveStats() const {
return interp_gain_curve_.get_stats();
}
void Limiter::SetSampleRate(size_t sample_rate_hz) {
CheckLimiterSampleRate(sample_rate_hz);
level_estimator_.SetSampleRate(sample_rate_hz);
}
void Limiter::Reset() {
level_estimator_.Reset();
}
float Limiter::LastAudioLevel() const {
return level_estimator_.LastAudioLevel();
}
} // namespace webrtc