mirror of
https://github.com/dolphin-emu/dolphin.git
synced 2024-11-14 21:37:52 -07:00
AudioCommon: extract AudioStretcher class for time-stretching
This commit is contained in:
parent
a4508e85e8
commit
87a467fe42
@ -38,6 +38,7 @@
|
||||
<ItemGroup>
|
||||
<ClCompile Include="aldlist.cpp" />
|
||||
<ClCompile Include="AudioCommon.cpp" />
|
||||
<ClCompile Include="AudioStretcher.cpp" />
|
||||
<ClCompile Include="CubebStream.cpp" />
|
||||
<ClCompile Include="CubebUtils.cpp" />
|
||||
<ClCompile Include="DPL2Decoder.cpp" />
|
||||
@ -54,6 +55,7 @@
|
||||
<ClInclude Include="aldlist.h" />
|
||||
<ClInclude Include="AlsaSoundStream.h" />
|
||||
<ClInclude Include="AudioCommon.h" />
|
||||
<ClInclude Include="AudioStretcher.h" />
|
||||
<ClInclude Include="CoreAudioSoundStream.h" />
|
||||
<ClInclude Include="CubebStream.h" />
|
||||
<ClInclude Include="CubebUtils.h" />
|
||||
|
@ -8,6 +8,7 @@
|
||||
<ItemGroup>
|
||||
<ClCompile Include="aldlist.cpp" />
|
||||
<ClCompile Include="AudioCommon.cpp" />
|
||||
<ClCompile Include="AudioStretcher.cpp" />
|
||||
<ClCompile Include="CubebUtils.cpp" />
|
||||
<ClCompile Include="DPL2Decoder.cpp" />
|
||||
<ClCompile Include="Mixer.cpp" />
|
||||
@ -31,6 +32,7 @@
|
||||
<ItemGroup>
|
||||
<ClInclude Include="aldlist.h" />
|
||||
<ClInclude Include="AudioCommon.h" />
|
||||
<ClInclude Include="AudioStretcher.h" />
|
||||
<ClInclude Include="CubebUtils.h" />
|
||||
<ClInclude Include="DPL2Decoder.h" />
|
||||
<ClInclude Include="Mixer.h" />
|
||||
|
85
Source/Core/AudioCommon/AudioStretcher.cpp
Normal file
85
Source/Core/AudioCommon/AudioStretcher.cpp
Normal file
@ -0,0 +1,85 @@
|
||||
// Copyright 2017 Dolphin Emulator Project
|
||||
// Licensed under GPLv2+
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
#include <cmath>
|
||||
|
||||
#include "AudioCommon/AudioStretcher.h"
|
||||
#include "Common/Logging/Log.h"
|
||||
#include "Core/ConfigManager.h"
|
||||
|
||||
namespace AudioCommon
|
||||
{
|
||||
AudioStretcher::AudioStretcher(unsigned int sample_rate) : m_sample_rate(sample_rate)
|
||||
{
|
||||
m_sound_touch.setChannels(2);
|
||||
m_sound_touch.setSampleRate(sample_rate);
|
||||
m_sound_touch.setPitch(1.0);
|
||||
m_sound_touch.setTempo(1.0);
|
||||
m_sound_touch.setSetting(SETTING_USE_QUICKSEEK, 0);
|
||||
m_sound_touch.setSetting(SETTING_SEQUENCE_MS, 62);
|
||||
m_sound_touch.setSetting(SETTING_SEEKWINDOW_MS, 28);
|
||||
m_sound_touch.setSetting(SETTING_OVERLAP_MS, 8);
|
||||
}
|
||||
|
||||
void AudioStretcher::Clear()
|
||||
{
|
||||
m_sound_touch.clear();
|
||||
}
|
||||
|
||||
void AudioStretcher::StretchAudio(const short* in, unsigned int num_in, short* out,
|
||||
unsigned int num_out)
|
||||
{
|
||||
const double time_delta = static_cast<double>(num_out) / m_sample_rate; // seconds
|
||||
|
||||
// We were given actual_samples number of samples, and num_samples were requested from us.
|
||||
double current_ratio = static_cast<double>(num_in) / static_cast<double>(num_out);
|
||||
|
||||
const double max_latency = SConfig::GetInstance().m_audio_stretch_max_latency;
|
||||
const double max_backlog = m_sample_rate * max_latency / 1000.0 / m_stretch_ratio;
|
||||
const double backlog_fullness = m_sound_touch.numSamples() / max_backlog;
|
||||
if (backlog_fullness > 5.0)
|
||||
{
|
||||
// Too many samples in backlog: Don't push anymore on
|
||||
num_in = 0;
|
||||
}
|
||||
|
||||
// We ideally want the backlog to be about 50% full.
|
||||
// This gives some headroom both ways to prevent underflow and overflow.
|
||||
// We tweak current_ratio to encourage this.
|
||||
constexpr double tweak_time_scale = 0.5; // seconds
|
||||
current_ratio *= 1.0 + 2.0 * (backlog_fullness - 0.5) * (time_delta / tweak_time_scale);
|
||||
|
||||
// This low-pass filter smoothes out variance in the calculated stretch ratio.
|
||||
// The time-scale determines how responsive this filter is.
|
||||
constexpr double lpf_time_scale = 1.0; // seconds
|
||||
const double lpf_gain = 1.0 - std::exp(-time_delta / lpf_time_scale);
|
||||
m_stretch_ratio += lpf_gain * (current_ratio - m_stretch_ratio);
|
||||
|
||||
// Place a lower limit of 10% speed. When a game boots up, there will be
|
||||
// many silence samples. These do not need to be timestretched.
|
||||
m_stretch_ratio = std::max(m_stretch_ratio, 0.1);
|
||||
m_sound_touch.setTempo(m_stretch_ratio);
|
||||
|
||||
DEBUG_LOG(AUDIO, "Audio stretching: samples:%u/%u ratio:%f backlog:%f gain: %f", num_in, num_out,
|
||||
m_stretch_ratio, backlog_fullness, lpf_gain);
|
||||
|
||||
m_sound_touch.putSamples(in, num_in);
|
||||
|
||||
const size_t samples_received = m_sound_touch.receiveSamples(out, num_out);
|
||||
|
||||
if (samples_received != 0)
|
||||
{
|
||||
m_last_stretched_sample[0] = out[samples_received * 2 - 2];
|
||||
m_last_stretched_sample[1] = out[samples_received * 2 - 1];
|
||||
}
|
||||
|
||||
// Perform padding if we've run out of samples.
|
||||
for (size_t i = samples_received; i < num_out; i++)
|
||||
{
|
||||
out[i * 2 + 0] = m_last_stretched_sample[0];
|
||||
out[i * 2 + 1] = m_last_stretched_sample[1];
|
||||
}
|
||||
}
|
||||
|
||||
} // namespace AudioCommon
|
27
Source/Core/AudioCommon/AudioStretcher.h
Normal file
27
Source/Core/AudioCommon/AudioStretcher.h
Normal file
@ -0,0 +1,27 @@
|
||||
// Copyright 2017 Dolphin Emulator Project
|
||||
// Licensed under GPLv2+
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <array>
|
||||
|
||||
#include <soundtouch/SoundTouch.h>
|
||||
|
||||
namespace AudioCommon
|
||||
{
|
||||
class AudioStretcher
|
||||
{
|
||||
public:
|
||||
AudioStretcher(unsigned int sample_rate);
|
||||
void StretchAudio(const short* in, unsigned int num_in, short* out, unsigned int num_out);
|
||||
void Clear();
|
||||
|
||||
private:
|
||||
unsigned int m_sample_rate;
|
||||
std::array<short, 2> m_last_stretched_sample = {};
|
||||
soundtouch::SoundTouch m_sound_touch;
|
||||
double m_stretch_ratio = 1.0;
|
||||
};
|
||||
|
||||
} // AudioCommon
|
@ -1,5 +1,6 @@
|
||||
set(SRCS
|
||||
AudioCommon.cpp
|
||||
AudioStretcher.cpp
|
||||
CubebStream.cpp
|
||||
CubebUtils.cpp
|
||||
DPL2Decoder.cpp
|
||||
|
@ -14,19 +14,10 @@
|
||||
#include "Common/Swap.h"
|
||||
#include "Core/ConfigManager.h"
|
||||
|
||||
CMixer::CMixer(unsigned int BackendSampleRate) : m_sampleRate(BackendSampleRate)
|
||||
CMixer::CMixer(unsigned int BackendSampleRate)
|
||||
: m_sampleRate(BackendSampleRate), m_stretcher(BackendSampleRate)
|
||||
{
|
||||
INFO_LOG(AUDIO_INTERFACE, "Mixer is initialized");
|
||||
|
||||
m_sound_touch.setChannels(2);
|
||||
m_sound_touch.setSampleRate(BackendSampleRate);
|
||||
m_sound_touch.setPitch(1.0);
|
||||
m_sound_touch.setTempo(1.0);
|
||||
m_sound_touch.setSetting(SETTING_USE_QUICKSEEK, 0);
|
||||
m_sound_touch.setSetting(SETTING_SEQUENCE_MS, 62);
|
||||
m_sound_touch.setSetting(SETTING_SEEKWINDOW_MS, 28);
|
||||
m_sound_touch.setSetting(SETTING_OVERLAP_MS, 8);
|
||||
|
||||
DPL2Reset();
|
||||
}
|
||||
|
||||
@ -146,10 +137,10 @@ unsigned int CMixer::Mix(short* samples, unsigned int num_samples)
|
||||
|
||||
if (!m_is_stretching)
|
||||
{
|
||||
m_sound_touch.clear();
|
||||
m_stretcher.Clear();
|
||||
m_is_stretching = true;
|
||||
}
|
||||
StretchAudio(m_stretch_buffer.data(), available_samples, samples, num_samples);
|
||||
m_stretcher.StretchAudio(m_stretch_buffer.data(), available_samples, samples, num_samples);
|
||||
}
|
||||
else
|
||||
{
|
||||
@ -181,60 +172,6 @@ unsigned int CMixer::MixSurround(float* samples, unsigned int num_samples)
|
||||
return available_samples;
|
||||
}
|
||||
|
||||
void CMixer::StretchAudio(const short* in, unsigned int num_in, short* out, unsigned int num_out)
|
||||
{
|
||||
const double time_delta = static_cast<double>(num_out) / m_sampleRate; // seconds
|
||||
|
||||
// We were given actual_samples number of samples, and num_samples were requested from us.
|
||||
double current_ratio = static_cast<double>(num_in) / static_cast<double>(num_out);
|
||||
|
||||
const double max_latency = SConfig::GetInstance().m_audio_stretch_max_latency;
|
||||
const double max_backlog = m_sampleRate * max_latency / 1000.0 / m_stretch_ratio;
|
||||
const double backlog_fullness = m_sound_touch.numSamples() / max_backlog;
|
||||
if (backlog_fullness > 5.0)
|
||||
{
|
||||
// Too many samples in backlog: Don't push anymore on
|
||||
num_in = 0;
|
||||
}
|
||||
|
||||
// We ideally want the backlog to be about 50% full.
|
||||
// This gives some headroom both ways to prevent underflow and overflow.
|
||||
// We tweak current_ratio to encourage this.
|
||||
constexpr double tweak_time_scale = 0.5; // seconds
|
||||
current_ratio *= 1.0 + 2.0 * (backlog_fullness - 0.5) * (time_delta / tweak_time_scale);
|
||||
|
||||
// This low-pass filter smoothes out variance in the calculated stretch ratio.
|
||||
// The time-scale determines how responsive this filter is.
|
||||
constexpr double lpf_time_scale = 1.0; // seconds
|
||||
const double m_lpf_gain = 1.0 - std::exp(-time_delta / lpf_time_scale);
|
||||
m_stretch_ratio += m_lpf_gain * (current_ratio - m_stretch_ratio);
|
||||
|
||||
// Place a lower limit of 10% speed. When a game boots up, there will be
|
||||
// many silence samples. These do not need to be timestretched.
|
||||
m_stretch_ratio = std::max(m_stretch_ratio, 0.1);
|
||||
m_sound_touch.setTempo(m_stretch_ratio);
|
||||
|
||||
DEBUG_LOG(AUDIO, "Audio stretching: samples:%u/%u ratio:%f backlog:%f gain: %f", num_in, num_out,
|
||||
m_stretch_ratio, backlog_fullness, m_lpf_gain);
|
||||
|
||||
m_sound_touch.putSamples(in, num_in);
|
||||
|
||||
const size_t samples_received = m_sound_touch.receiveSamples(out, num_out);
|
||||
|
||||
if (samples_received != 0)
|
||||
{
|
||||
m_last_stretched_sample[0] = out[samples_received * 2 - 2];
|
||||
m_last_stretched_sample[1] = out[samples_received * 2 - 1];
|
||||
}
|
||||
|
||||
// Preform padding if we've run out of samples.
|
||||
for (size_t i = samples_received; i < num_out; i++)
|
||||
{
|
||||
out[i * 2 + 0] = m_last_stretched_sample[0];
|
||||
out[i * 2 + 1] = m_last_stretched_sample[1];
|
||||
}
|
||||
}
|
||||
|
||||
void CMixer::MixerFifo::PushSamples(const short* samples, unsigned int num_samples)
|
||||
{
|
||||
// Cache access in non-volatile variable
|
||||
|
@ -7,12 +7,10 @@
|
||||
#include <array>
|
||||
#include <atomic>
|
||||
|
||||
#include "AudioCommon/AudioStretcher.h"
|
||||
#include "AudioCommon/WaveFile.h"
|
||||
#include "Common/CommonTypes.h"
|
||||
|
||||
#include <soundtouch/STTypes.h>
|
||||
#include <soundtouch/SoundTouch.h>
|
||||
|
||||
class CMixer final
|
||||
{
|
||||
public:
|
||||
@ -76,17 +74,13 @@ private:
|
||||
u32 m_frac = 0;
|
||||
};
|
||||
|
||||
void StretchAudio(const short* in, unsigned int num_in, short* out, unsigned int num_out);
|
||||
|
||||
MixerFifo m_dma_mixer{this, 32000};
|
||||
MixerFifo m_streaming_mixer{this, 48000};
|
||||
MixerFifo m_wiimote_speaker_mixer{this, 3000};
|
||||
unsigned int m_sampleRate;
|
||||
|
||||
bool m_is_stretching = false;
|
||||
soundtouch::SoundTouch m_sound_touch;
|
||||
double m_stretch_ratio = 1.0;
|
||||
std::array<short, 2> m_last_stretched_sample = {};
|
||||
AudioCommon::AudioStretcher m_stretcher;
|
||||
std::array<short, MAX_SAMPLES * 2> m_stretch_buffer;
|
||||
std::array<float, MAX_SAMPLES * 2> m_float_conversion_buffer;
|
||||
|
||||
|
Loading…
Reference in New Issue
Block a user