mirror of
https://github.com/PabloMK7/citra.git
synced 2025-10-31 05:40:04 +00:00
Merge pull request #4194 from MerryMage/audiofifo
audio_core: Simplify sink interface
This commit is contained in:
Merry
GitHub
commit
bb9e92c77c
12 changed files with 316 additions and 314 deletions
|
|
@ -2,6 +2,7 @@
|
|||
// Licensed under GPLv2 or any later version
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
#include <cstdarg>
|
||||
#include <mutex>
|
||||
#include <vector>
|
||||
#include <cubeb/cubeb.h>
|
||||
|
|
@ -13,17 +14,16 @@ namespace AudioCore {
|
|||
|
||||
struct CubebSink::Impl {
|
||||
unsigned int sample_rate = 0;
|
||||
std::vector<std::string> device_list;
|
||||
|
||||
cubeb* ctx = nullptr;
|
||||
cubeb_stream* stream = nullptr;
|
||||
|
||||
std::mutex queue_mutex;
|
||||
std::vector<s16> queue;
|
||||
std::function<void(s16*, std::size_t)> cb;
|
||||
|
||||
static long DataCallback(cubeb_stream* stream, void* user_data, const void* input_buffer,
|
||||
void* output_buffer, long num_frames);
|
||||
static void StateCallback(cubeb_stream* stream, void* user_data, cubeb_state state);
|
||||
static void LogCallback(char const* fmt, ...);
|
||||
};
|
||||
|
||||
CubebSink::CubebSink(std::string target_device_name) : impl(std::make_unique<Impl>()) {
|
||||
|
|
@ -31,21 +31,23 @@ CubebSink::CubebSink(std::string target_device_name) : impl(std::make_unique<Imp
|
|||
LOG_CRITICAL(Audio_Sink, "cubeb_init failed");
|
||||
return;
|
||||
}
|
||||
|
||||
cubeb_devid output_device = nullptr;
|
||||
|
||||
cubeb_stream_params params;
|
||||
params.rate = native_sample_rate;
|
||||
params.channels = 2;
|
||||
params.format = CUBEB_SAMPLE_S16NE;
|
||||
params.layout = CUBEB_LAYOUT_STEREO;
|
||||
cubeb_set_log_callback(CUBEB_LOG_NORMAL, &Impl::LogCallback);
|
||||
|
||||
impl->sample_rate = native_sample_rate;
|
||||
|
||||
u32 minimum_latency = 0;
|
||||
if (cubeb_get_min_latency(impl->ctx, ¶ms, &minimum_latency) != CUBEB_OK)
|
||||
LOG_CRITICAL(Audio_Sink, "Error getting minimum latency");
|
||||
cubeb_stream_params params;
|
||||
params.rate = impl->sample_rate;
|
||||
params.channels = 2;
|
||||
params.layout = CUBEB_LAYOUT_STEREO;
|
||||
params.format = CUBEB_SAMPLE_S16NE;
|
||||
params.prefs = CUBEB_STREAM_PREF_NONE;
|
||||
|
||||
u32 minimum_latency = 100 * impl->sample_rate / 1000; // Firefox default
|
||||
if (cubeb_get_min_latency(impl->ctx, ¶ms, &minimum_latency) != CUBEB_OK) {
|
||||
LOG_CRITICAL(Audio_Sink, "Error getting minimum latency");
|
||||
}
|
||||
|
||||
cubeb_devid output_device = nullptr;
|
||||
if (target_device_name != auto_device_name && !target_device_name.empty()) {
|
||||
cubeb_device_collection collection;
|
||||
if (cubeb_enumerate_devices(impl->ctx, CUBEB_DEVICE_TYPE_OUTPUT, &collection) != CUBEB_OK) {
|
||||
|
|
@ -63,10 +65,22 @@ CubebSink::CubebSink(std::string target_device_name) : impl(std::make_unique<Imp
|
|||
}
|
||||
}
|
||||
|
||||
if (cubeb_stream_init(impl->ctx, &impl->stream, "Citra Audio Output", nullptr, nullptr,
|
||||
output_device, ¶ms, std::max(512u, minimum_latency),
|
||||
&Impl::DataCallback, &Impl::StateCallback, impl.get()) != CUBEB_OK) {
|
||||
LOG_CRITICAL(Audio_Sink, "Error initializing cubeb stream");
|
||||
int stream_err = cubeb_stream_init(impl->ctx, &impl->stream, "CitraAudio", nullptr, nullptr,
|
||||
output_device, ¶ms, std::max(512u, minimum_latency),
|
||||
&Impl::DataCallback, &Impl::StateCallback, impl.get());
|
||||
if (stream_err != CUBEB_OK) {
|
||||
switch (stream_err) {
|
||||
case CUBEB_ERROR:
|
||||
default:
|
||||
LOG_CRITICAL(Audio_Sink, "Error initializing cubeb stream ({})", stream_err);
|
||||
break;
|
||||
case CUBEB_ERROR_INVALID_FORMAT:
|
||||
LOG_CRITICAL(Audio_Sink, "Invalid format when initializing cubeb stream");
|
||||
break;
|
||||
case CUBEB_ERROR_DEVICE_UNAVAILABLE:
|
||||
LOG_CRITICAL(Audio_Sink, "Device unavailable when initializing cubeb stream");
|
||||
break;
|
||||
}
|
||||
return;
|
||||
}
|
||||
|
||||
|
|
@ -77,8 +91,11 @@ CubebSink::CubebSink(std::string target_device_name) : impl(std::make_unique<Imp
|
|||
}
|
||||
|
||||
CubebSink::~CubebSink() {
|
||||
if (!impl->ctx)
|
||||
if (!impl->ctx) {
|
||||
return;
|
||||
}
|
||||
|
||||
impl->cb = nullptr;
|
||||
|
||||
if (cubeb_stream_stop(impl->stream) != CUBEB_OK) {
|
||||
LOG_CRITICAL(Audio_Sink, "Error stopping cubeb stream");
|
||||
|
|
@ -95,56 +112,62 @@ unsigned int CubebSink::GetNativeSampleRate() const {
|
|||
return impl->sample_rate;
|
||||
}
|
||||
|
||||
void CubebSink::EnqueueSamples(const s16* samples, std::size_t sample_count) {
|
||||
if (!impl->ctx)
|
||||
return;
|
||||
|
||||
std::lock_guard lock{impl->queue_mutex};
|
||||
|
||||
impl->queue.reserve(impl->queue.size() + sample_count * 2);
|
||||
std::copy(samples, samples + sample_count * 2, std::back_inserter(impl->queue));
|
||||
}
|
||||
|
||||
size_t CubebSink::SamplesInQueue() const {
|
||||
if (!impl->ctx)
|
||||
return 0;
|
||||
|
||||
std::lock_guard lock{impl->queue_mutex};
|
||||
return impl->queue.size() / 2;
|
||||
void CubebSink::SetCallback(std::function<void(s16*, std::size_t)> cb) {
|
||||
impl->cb = cb;
|
||||
}
|
||||
|
||||
long CubebSink::Impl::DataCallback(cubeb_stream* stream, void* user_data, const void* input_buffer,
|
||||
void* output_buffer, long num_frames) {
|
||||
Impl* impl = static_cast<Impl*>(user_data);
|
||||
u8* buffer = reinterpret_cast<u8*>(output_buffer);
|
||||
s16* buffer = reinterpret_cast<s16*>(output_buffer);
|
||||
|
||||
if (!impl)
|
||||
return 0;
|
||||
|
||||
std::lock_guard lock{impl->queue_mutex};
|
||||
|
||||
std::size_t frames_to_write =
|
||||
std::min(impl->queue.size() / 2, static_cast<std::size_t>(num_frames));
|
||||
|
||||
memcpy(buffer, impl->queue.data(), frames_to_write * sizeof(s16) * 2);
|
||||
impl->queue.erase(impl->queue.begin(), impl->queue.begin() + frames_to_write * 2);
|
||||
|
||||
if (frames_to_write < num_frames) {
|
||||
// Fill the rest of the frames with silence
|
||||
memset(buffer + frames_to_write * sizeof(s16) * 2, 0,
|
||||
(num_frames - frames_to_write) * sizeof(s16) * 2);
|
||||
if (!impl || !impl->cb) {
|
||||
LOG_DEBUG(Audio_Sink, "Emitting zeros");
|
||||
std::memset(output_buffer, 0, num_frames * 2 * sizeof(s16));
|
||||
return num_frames;
|
||||
}
|
||||
|
||||
impl->cb(buffer, num_frames);
|
||||
|
||||
return num_frames;
|
||||
}
|
||||
|
||||
void CubebSink::Impl::StateCallback(cubeb_stream* stream, void* user_data, cubeb_state state) {}
|
||||
void CubebSink::Impl::StateCallback(cubeb_stream* stream, void* user_data, cubeb_state state) {
|
||||
switch (state) {
|
||||
case CUBEB_STATE_STARTED:
|
||||
LOG_INFO(Audio_Sink, "Cubeb Audio Stream Started");
|
||||
break;
|
||||
case CUBEB_STATE_STOPPED:
|
||||
LOG_INFO(Audio_Sink, "Cubeb Audio Stream Stopped");
|
||||
break;
|
||||
case CUBEB_STATE_DRAINED:
|
||||
LOG_INFO(Audio_Sink, "Cubeb Audio Stream Drained");
|
||||
break;
|
||||
case CUBEB_STATE_ERROR:
|
||||
LOG_CRITICAL(Audio_Sink, "Cubeb Audio Stream Errored");
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
void CubebSink::Impl::LogCallback(char const* format, ...) {
|
||||
std::array<char, 512> buffer;
|
||||
std::va_list args;
|
||||
va_start(args, format);
|
||||
#ifdef _MSC_VER
|
||||
vsprintf_s(buffer.data(), buffer.size(), format, args);
|
||||
#else
|
||||
vsnprintf(buffer.data(), buffer.size(), format, args);
|
||||
#endif
|
||||
va_end(args);
|
||||
buffer.back() = '\0';
|
||||
LOG_INFO(Audio_Sink, "{}", buffer.data());
|
||||
}
|
||||
|
||||
std::vector<std::string> ListCubebSinkDevices() {
|
||||
std::vector<std::string> device_list;
|
||||
cubeb* ctx;
|
||||
|
||||
if (cubeb_init(&ctx, "Citra Device Enumerator", nullptr) != CUBEB_OK) {
|
||||
if (cubeb_init(&ctx, "CitraEnumerator", nullptr) != CUBEB_OK) {
|
||||
LOG_CRITICAL(Audio_Sink, "cubeb_init failed");
|
||||
return {};
|
||||
}
|
||||
|
|
|
|||
|
|
@ -17,9 +17,7 @@ public:
|
|||
|
||||
unsigned int GetNativeSampleRate() const override;
|
||||
|
||||
void EnqueueSamples(const s16* samples, std::size_t sample_count) override;
|
||||
|
||||
std::size_t SamplesInQueue() const override;
|
||||
void SetCallback(std::function<void(s16*, std::size_t)> cb) override;
|
||||
|
||||
private:
|
||||
struct Impl;
|
||||
|
|
|
|||
|
|
@ -12,16 +12,13 @@
|
|||
namespace AudioCore {
|
||||
|
||||
DspInterface::DspInterface() = default;
|
||||
|
||||
DspInterface::~DspInterface() {
|
||||
if (perform_time_stretching) {
|
||||
FlushResidualStretcherAudio();
|
||||
}
|
||||
}
|
||||
DspInterface::~DspInterface() = default;
|
||||
|
||||
void DspInterface::SetSink(const std::string& sink_id, const std::string& audio_device) {
|
||||
const SinkDetails& sink_details = GetSinkDetails(sink_id);
|
||||
sink = sink_details.factory(audio_device);
|
||||
sink->SetCallback(
|
||||
[this](s16* buffer, std::size_t num_frames) { OutputCallback(buffer, num_frames); });
|
||||
time_stretcher.SetOutputSampleRate(sink->GetNativeSampleRate());
|
||||
}
|
||||
|
||||
|
|
@ -35,7 +32,7 @@ void DspInterface::EnableStretching(bool enable) {
|
|||
return;
|
||||
|
||||
if (!enable) {
|
||||
FlushResidualStretcherAudio();
|
||||
flushing_time_stretcher = true;
|
||||
}
|
||||
perform_time_stretching = enable;
|
||||
}
|
||||
|
|
@ -44,39 +41,41 @@ void DspInterface::OutputFrame(StereoFrame16& frame) {
|
|||
if (!sink)
|
||||
return;
|
||||
|
||||
// Implementation of the hardware volume slider with a dynamic range of 60 dB
|
||||
double volume_scale_factor = std::exp(6.90775 * Settings::values.volume) * 0.001;
|
||||
for (std::size_t i = 0; i < frame.size(); i++) {
|
||||
frame[i][0] = static_cast<s16>(frame[i][0] * volume_scale_factor);
|
||||
frame[i][1] = static_cast<s16>(frame[i][1] * volume_scale_factor);
|
||||
}
|
||||
|
||||
if (perform_time_stretching) {
|
||||
time_stretcher.AddSamples(&frame[0][0], frame.size());
|
||||
std::vector<s16> stretched_samples = time_stretcher.Process(sink->SamplesInQueue());
|
||||
sink->EnqueueSamples(stretched_samples.data(), stretched_samples.size() / 2);
|
||||
} else {
|
||||
constexpr std::size_t maximum_sample_latency = 2048; // about 64 miliseconds
|
||||
if (sink->SamplesInQueue() > maximum_sample_latency) {
|
||||
// This can occur if we're running too fast and samples are starting to back up.
|
||||
// Just drop the samples.
|
||||
return;
|
||||
}
|
||||
|
||||
sink->EnqueueSamples(&frame[0][0], frame.size());
|
||||
}
|
||||
fifo.Push(frame.data(), frame.size());
|
||||
}
|
||||
|
||||
void DspInterface::FlushResidualStretcherAudio() {
|
||||
if (!sink)
|
||||
return;
|
||||
void DspInterface::OutputCallback(s16* buffer, std::size_t num_frames) {
|
||||
std::size_t frames_written;
|
||||
if (perform_time_stretching) {
|
||||
const std::vector<s16> in{fifo.Pop()};
|
||||
const std::size_t num_in{in.size() / 2};
|
||||
frames_written = time_stretcher.Process(in.data(), num_in, buffer, num_frames);
|
||||
} else if (flushing_time_stretcher) {
|
||||
time_stretcher.Flush();
|
||||
frames_written = time_stretcher.Process(nullptr, 0, buffer, num_frames);
|
||||
frames_written += fifo.Pop(buffer, num_frames - frames_written);
|
||||
flushing_time_stretcher = false;
|
||||
} else {
|
||||
frames_written = fifo.Pop(buffer, num_frames);
|
||||
}
|
||||
|
||||
time_stretcher.Flush();
|
||||
while (true) {
|
||||
std::vector<s16> residual_audio = time_stretcher.Process(sink->SamplesInQueue());
|
||||
if (residual_audio.empty())
|
||||
break;
|
||||
sink->EnqueueSamples(residual_audio.data(), residual_audio.size() / 2);
|
||||
if (frames_written > 0) {
|
||||
std::memcpy(&last_frame[0], buffer + 2 * (frames_written - 1), 2 * sizeof(s16));
|
||||
}
|
||||
|
||||
// Hold last emitted frame; this prevents popping.
|
||||
for (std::size_t i = frames_written; i < num_frames; i++) {
|
||||
std::memcpy(buffer + 2 * i, &last_frame[0], 2 * sizeof(s16));
|
||||
}
|
||||
|
||||
// Implementation of the hardware volume slider with a dynamic range of 60 dB
|
||||
const float linear_volume = std::clamp(Settings::values.volume, 0.0f, 1.0f);
|
||||
if (linear_volume != 1.0) {
|
||||
const float volume_scale_factor = std::exp(6.90775f * linear_volume) * 0.001f;
|
||||
for (std::size_t i = 0; i < num_frames; i++) {
|
||||
buffer[i * 2 + 0] = static_cast<s16>(buffer[i * 2 + 0] * volume_scale_factor);
|
||||
buffer[i * 2 + 1] = static_cast<s16>(buffer[i * 2 + 1] * volume_scale_factor);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -9,6 +9,7 @@
|
|||
#include "audio_core/audio_types.h"
|
||||
#include "audio_core/time_stretch.h"
|
||||
#include "common/common_types.h"
|
||||
#include "common/ring_buffer.h"
|
||||
#include "core/memory.h"
|
||||
|
||||
namespace Service {
|
||||
|
|
@ -81,9 +82,13 @@ protected:
|
|||
|
||||
private:
|
||||
void FlushResidualStretcherAudio();
|
||||
void OutputCallback(s16* buffer, std::size_t num_frames);
|
||||
|
||||
std::unique_ptr<Sink> sink;
|
||||
bool perform_time_stretching = false;
|
||||
std::atomic<bool> perform_time_stretching = false;
|
||||
std::atomic<bool> flushing_time_stretcher = false;
|
||||
Common::RingBuffer<s16, 0x2000, 2> fifo;
|
||||
std::array<s16, 2> last_frame{};
|
||||
TimeStretcher time_stretcher;
|
||||
};
|
||||
|
||||
|
|
|
|||
|
|
@ -19,11 +19,7 @@ public:
|
|||
return native_sample_rate;
|
||||
}
|
||||
|
||||
void EnqueueSamples(const s16*, std::size_t) override {}
|
||||
|
||||
std::size_t SamplesInQueue() const override {
|
||||
return 0;
|
||||
}
|
||||
void SetCallback(std::function<void(s16*, std::size_t)>) override {}
|
||||
};
|
||||
|
||||
} // namespace AudioCore
|
||||
|
|
|
|||
|
|
@ -2,8 +2,8 @@
|
|||
// Licensed under GPLv2 or any later version
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
#include <list>
|
||||
#include <numeric>
|
||||
#include <string>
|
||||
#include <vector>
|
||||
#include <SDL.h>
|
||||
#include "audio_core/audio_types.h"
|
||||
#include "audio_core/sdl2_sink.h"
|
||||
|
|
@ -17,7 +17,7 @@ struct SDL2Sink::Impl {
|
|||
|
||||
SDL_AudioDeviceID audio_device_id = 0;
|
||||
|
||||
std::list<std::vector<s16>> queue;
|
||||
std::function<void(s16*, std::size_t)> cb;
|
||||
|
||||
static void Callback(void* impl_, u8* buffer, int buffer_size_in_bytes);
|
||||
};
|
||||
|
|
@ -74,58 +74,18 @@ unsigned int SDL2Sink::GetNativeSampleRate() const {
|
|||
return impl->sample_rate;
|
||||
}
|
||||
|
||||
void SDL2Sink::EnqueueSamples(const s16* samples, std::size_t sample_count) {
|
||||
if (impl->audio_device_id <= 0)
|
||||
return;
|
||||
|
||||
SDL_LockAudioDevice(impl->audio_device_id);
|
||||
impl->queue.emplace_back(samples, samples + sample_count * 2);
|
||||
SDL_UnlockAudioDevice(impl->audio_device_id);
|
||||
}
|
||||
|
||||
size_t SDL2Sink::SamplesInQueue() const {
|
||||
if (impl->audio_device_id <= 0)
|
||||
return 0;
|
||||
|
||||
SDL_LockAudioDevice(impl->audio_device_id);
|
||||
|
||||
std::size_t total_size =
|
||||
std::accumulate(impl->queue.begin(), impl->queue.end(), static_cast<std::size_t>(0),
|
||||
[](std::size_t sum, const auto& buffer) {
|
||||
// Division by two because each stereo sample is made of
|
||||
// two s16.
|
||||
return sum + buffer.size() / 2;
|
||||
});
|
||||
|
||||
SDL_UnlockAudioDevice(impl->audio_device_id);
|
||||
|
||||
return total_size;
|
||||
void SDL2Sink::SetCallback(std::function<void(s16*, std::size_t)> cb) {
|
||||
impl->cb = cb;
|
||||
}
|
||||
|
||||
void SDL2Sink::Impl::Callback(void* impl_, u8* buffer, int buffer_size_in_bytes) {
|
||||
Impl* impl = reinterpret_cast<Impl*>(impl_);
|
||||
if (!impl || !impl->cb)
|
||||
return;
|
||||
|
||||
std::size_t remaining_size = static_cast<std::size_t>(buffer_size_in_bytes) /
|
||||
sizeof(s16); // Keep track of size in 16-bit increments.
|
||||
const size_t num_frames = buffer_size_in_bytes / (2 * sizeof(s16));
|
||||
|
||||
while (remaining_size > 0 && !impl->queue.empty()) {
|
||||
if (impl->queue.front().size() <= remaining_size) {
|
||||
memcpy(buffer, impl->queue.front().data(), impl->queue.front().size() * sizeof(s16));
|
||||
buffer += impl->queue.front().size() * sizeof(s16);
|
||||
remaining_size -= impl->queue.front().size();
|
||||
impl->queue.pop_front();
|
||||
} else {
|
||||
memcpy(buffer, impl->queue.front().data(), remaining_size * sizeof(s16));
|
||||
buffer += remaining_size * sizeof(s16);
|
||||
impl->queue.front().erase(impl->queue.front().begin(),
|
||||
impl->queue.front().begin() + remaining_size);
|
||||
remaining_size = 0;
|
||||
}
|
||||
}
|
||||
|
||||
if (remaining_size > 0) {
|
||||
memset(buffer, 0, remaining_size * sizeof(s16));
|
||||
}
|
||||
impl->cb(reinterpret_cast<s16*>(buffer), num_frames);
|
||||
}
|
||||
|
||||
std::vector<std::string> ListSDL2SinkDevices() {
|
||||
|
|
|
|||
|
|
@ -17,9 +17,7 @@ public:
|
|||
|
||||
unsigned int GetNativeSampleRate() const override;
|
||||
|
||||
void EnqueueSamples(const s16* samples, std::size_t sample_count) override;
|
||||
|
||||
std::size_t SamplesInQueue() const override;
|
||||
void SetCallback(std::function<void(s16*, std::size_t)> cb) override;
|
||||
|
||||
private:
|
||||
struct Impl;
|
||||
|
|
|
|||
|
|
@ -4,7 +4,7 @@
|
|||
|
||||
#pragma once
|
||||
|
||||
#include <vector>
|
||||
#include <functional>
|
||||
#include "common/common_types.h"
|
||||
|
||||
namespace AudioCore {
|
||||
|
|
@ -20,19 +20,16 @@ class Sink {
|
|||
public:
|
||||
virtual ~Sink() = default;
|
||||
|
||||
/// The native rate of this sink. The sink expects to be fed samples that respect this. (Units:
|
||||
/// samples/sec)
|
||||
/// The native rate of this sink. The sink expects to be fed samples that respect this.
|
||||
/// (Units: samples/sec)
|
||||
virtual unsigned int GetNativeSampleRate() const = 0;
|
||||
|
||||
/**
|
||||
* Feed stereo samples to sink.
|
||||
* Set callback for samples
|
||||
* @param samples Samples in interleaved stereo PCM16 format.
|
||||
* @param sample_count Number of samples.
|
||||
*/
|
||||
virtual void EnqueueSamples(const s16* samples, std::size_t sample_count) = 0;
|
||||
|
||||
/// Samples enqueued that have not been played yet.
|
||||
virtual std::size_t SamplesInQueue() const = 0;
|
||||
virtual void SetCallback(std::function<void(s16*, std::size_t)> cb) = 0;
|
||||
};
|
||||
|
||||
} // namespace AudioCore
|
||||
|
|
|
|||
|
|
@ -3,143 +3,75 @@
|
|||
// Refer to the license.txt file included.
|
||||
|
||||
#include <algorithm>
|
||||
#include <chrono>
|
||||
#include <cmath>
|
||||
#include <vector>
|
||||
#include <cstddef>
|
||||
#include <memory>
|
||||
#include <SoundTouch.h>
|
||||
#include "audio_core/audio_types.h"
|
||||
#include "audio_core/time_stretch.h"
|
||||
#include "common/common_types.h"
|
||||
#include "common/logging/log.h"
|
||||
|
||||
using steady_clock = std::chrono::steady_clock;
|
||||
|
||||
namespace AudioCore {
|
||||
|
||||
constexpr double MIN_RATIO = 0.1;
|
||||
constexpr double MAX_RATIO = 100.0;
|
||||
|
||||
static double ClampRatio(double ratio) {
|
||||
return std::clamp(ratio, MIN_RATIO, MAX_RATIO);
|
||||
TimeStretcher::TimeStretcher()
|
||||
: sample_rate(native_sample_rate), sound_touch(std::make_unique<soundtouch::SoundTouch>()) {
|
||||
sound_touch->setChannels(2);
|
||||
sound_touch->setSampleRate(native_sample_rate);
|
||||
sound_touch->setPitch(1.0);
|
||||
sound_touch->setTempo(1.0);
|
||||
}
|
||||
|
||||
constexpr double MIN_DELAY_TIME = 0.05; // Units: seconds
|
||||
constexpr double MAX_DELAY_TIME = 0.25; // Units: seconds
|
||||
constexpr std::size_t DROP_FRAMES_SAMPLE_DELAY = 16000; // Units: samples
|
||||
|
||||
constexpr double SMOOTHING_FACTOR = 0.007;
|
||||
|
||||
struct TimeStretcher::Impl {
|
||||
soundtouch::SoundTouch soundtouch;
|
||||
|
||||
steady_clock::time_point frame_timer = steady_clock::now();
|
||||
std::size_t samples_queued = 0;
|
||||
|
||||
double smoothed_ratio = 1.0;
|
||||
|
||||
double sample_rate = static_cast<double>(native_sample_rate);
|
||||
};
|
||||
|
||||
std::vector<s16> TimeStretcher::Process(std::size_t samples_in_queue) {
|
||||
// This is a very simple algorithm without any fancy control theory. It works and is stable.
|
||||
|
||||
double ratio = CalculateCurrentRatio();
|
||||
ratio = CorrectForUnderAndOverflow(ratio, samples_in_queue);
|
||||
impl->smoothed_ratio =
|
||||
(1.0 - SMOOTHING_FACTOR) * impl->smoothed_ratio + SMOOTHING_FACTOR * ratio;
|
||||
impl->smoothed_ratio = ClampRatio(impl->smoothed_ratio);
|
||||
|
||||
// SoundTouch's tempo definition the inverse of our ratio definition.
|
||||
impl->soundtouch.setTempo(1.0 / impl->smoothed_ratio);
|
||||
|
||||
std::vector<s16> samples = GetSamples();
|
||||
if (samples_in_queue >= DROP_FRAMES_SAMPLE_DELAY) {
|
||||
samples.clear();
|
||||
LOG_DEBUG(Audio, "Dropping frames!");
|
||||
}
|
||||
return samples;
|
||||
}
|
||||
|
||||
TimeStretcher::TimeStretcher() : impl(std::make_unique<Impl>()) {
|
||||
impl->soundtouch.setPitch(1.0);
|
||||
impl->soundtouch.setChannels(2);
|
||||
impl->soundtouch.setSampleRate(native_sample_rate);
|
||||
Reset();
|
||||
}
|
||||
|
||||
TimeStretcher::~TimeStretcher() {
|
||||
impl->soundtouch.clear();
|
||||
}
|
||||
TimeStretcher::~TimeStretcher() = default;
|
||||
|
||||
void TimeStretcher::SetOutputSampleRate(unsigned int sample_rate) {
|
||||
impl->sample_rate = static_cast<double>(sample_rate);
|
||||
impl->soundtouch.setRate(static_cast<double>(native_sample_rate) / impl->sample_rate);
|
||||
sound_touch->setSampleRate(sample_rate);
|
||||
sample_rate = native_sample_rate;
|
||||
}
|
||||
|
||||
void TimeStretcher::AddSamples(const s16* buffer, std::size_t num_samples) {
|
||||
impl->soundtouch.putSamples(buffer, static_cast<uint>(num_samples));
|
||||
impl->samples_queued += num_samples;
|
||||
std::size_t TimeStretcher::Process(const s16* in, std::size_t num_in, s16* out,
|
||||
std::size_t num_out) {
|
||||
const double time_delta = static_cast<double>(num_out) / sample_rate; // seconds
|
||||
double current_ratio = static_cast<double>(num_in) / static_cast<double>(num_out);
|
||||
|
||||
const double max_latency = 0.25; // seconds
|
||||
const double max_backlog = sample_rate * max_latency;
|
||||
const double backlog_fullness = sound_touch->numSamples() / max_backlog;
|
||||
if (backlog_fullness > 4.0) {
|
||||
// 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 = 0.050; // seconds
|
||||
const double tweak_correction = (backlog_fullness - 0.5) * (time_delta / tweak_time_scale);
|
||||
current_ratio *= std::pow(1.0 + 2.0 * tweak_correction, tweak_correction < 0 ? 3.0 : 1.0);
|
||||
|
||||
// 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 = 0.712; // seconds
|
||||
const double lpf_gain = 1.0 - std::exp(-time_delta / lpf_time_scale);
|
||||
stretch_ratio += lpf_gain * (current_ratio - stretch_ratio);
|
||||
|
||||
// Place a lower limit of 5% speed. When a game boots up, there will be
|
||||
// many silence samples. These do not need to be timestretched.
|
||||
stretch_ratio = std::max(stretch_ratio, 0.05);
|
||||
sound_touch->setTempo(stretch_ratio);
|
||||
|
||||
LOG_DEBUG(Audio, "{:5}/{:5} ratio:{:0.6f} backlog:{:0.6f}", num_in, num_out, stretch_ratio,
|
||||
backlog_fullness);
|
||||
|
||||
sound_touch->putSamples(in, num_in);
|
||||
return sound_touch->receiveSamples(out, num_out);
|
||||
}
|
||||
|
||||
void TimeStretcher::Clear() {
|
||||
sound_touch->clear();
|
||||
}
|
||||
|
||||
void TimeStretcher::Flush() {
|
||||
impl->soundtouch.flush();
|
||||
}
|
||||
|
||||
void TimeStretcher::Reset() {
|
||||
impl->soundtouch.setTempo(1.0);
|
||||
impl->soundtouch.clear();
|
||||
impl->smoothed_ratio = 1.0;
|
||||
impl->frame_timer = steady_clock::now();
|
||||
impl->samples_queued = 0;
|
||||
SetOutputSampleRate(native_sample_rate);
|
||||
}
|
||||
|
||||
double TimeStretcher::CalculateCurrentRatio() {
|
||||
const steady_clock::time_point now = steady_clock::now();
|
||||
const std::chrono::duration<double> duration = now - impl->frame_timer;
|
||||
|
||||
const double expected_time =
|
||||
static_cast<double>(impl->samples_queued) / static_cast<double>(native_sample_rate);
|
||||
const double actual_time = duration.count();
|
||||
|
||||
double ratio;
|
||||
if (expected_time != 0) {
|
||||
ratio = ClampRatio(actual_time / expected_time);
|
||||
} else {
|
||||
ratio = impl->smoothed_ratio;
|
||||
}
|
||||
|
||||
impl->frame_timer = now;
|
||||
impl->samples_queued = 0;
|
||||
|
||||
return ratio;
|
||||
}
|
||||
|
||||
double TimeStretcher::CorrectForUnderAndOverflow(double ratio, std::size_t sample_delay) const {
|
||||
const std::size_t min_sample_delay =
|
||||
static_cast<std::size_t>(MIN_DELAY_TIME * impl->sample_rate);
|
||||
const std::size_t max_sample_delay =
|
||||
static_cast<std::size_t>(MAX_DELAY_TIME * impl->sample_rate);
|
||||
|
||||
if (sample_delay < min_sample_delay) {
|
||||
// Make the ratio bigger.
|
||||
ratio = ratio > 1.0 ? ratio * ratio : sqrt(ratio);
|
||||
} else if (sample_delay > max_sample_delay) {
|
||||
// Make the ratio smaller.
|
||||
ratio = ratio > 1.0 ? sqrt(ratio) : ratio * ratio;
|
||||
}
|
||||
|
||||
return ClampRatio(ratio);
|
||||
}
|
||||
|
||||
std::vector<s16> TimeStretcher::GetSamples() {
|
||||
uint available = impl->soundtouch.numSamples();
|
||||
|
||||
std::vector<s16> output(static_cast<std::size_t>(available) * 2);
|
||||
|
||||
impl->soundtouch.receiveSamples(output.data(), available);
|
||||
|
||||
return output;
|
||||
sound_touch->flush();
|
||||
}
|
||||
|
||||
} // namespace AudioCore
|
||||
|
|
|
|||
|
|
@ -4,57 +4,39 @@
|
|||
|
||||
#pragma once
|
||||
|
||||
#include <array>
|
||||
#include <cstddef>
|
||||
#include <memory>
|
||||
#include <vector>
|
||||
#include "common/common_types.h"
|
||||
|
||||
namespace soundtouch {
|
||||
class SoundTouch;
|
||||
}
|
||||
|
||||
namespace AudioCore {
|
||||
|
||||
class TimeStretcher final {
|
||||
class TimeStretcher {
|
||||
public:
|
||||
TimeStretcher();
|
||||
~TimeStretcher();
|
||||
|
||||
/**
|
||||
* Set sample rate for the samples that Process returns.
|
||||
* @param sample_rate The sample rate.
|
||||
*/
|
||||
void SetOutputSampleRate(unsigned int sample_rate);
|
||||
|
||||
/**
|
||||
* Add samples to be processed.
|
||||
* @param sample_buffer Buffer of samples in interleaved stereo PCM16 format.
|
||||
* @param num_samples Number of samples.
|
||||
*/
|
||||
void AddSamples(const s16* sample_buffer, std::size_t num_samples);
|
||||
/// @param in Input sample buffer
|
||||
/// @param num_in Number of input frames in `in`
|
||||
/// @param out Output sample buffer
|
||||
/// @param num_out Desired number of output frames in `out`
|
||||
/// @returns Actual number of frames written to `out`
|
||||
std::size_t Process(const s16* in, std::size_t num_in, s16* out, std::size_t num_out);
|
||||
|
||||
void Clear();
|
||||
|
||||
/// Flush audio remaining in internal buffers.
|
||||
void Flush();
|
||||
|
||||
/// Resets internal state and clears buffers.
|
||||
void Reset();
|
||||
|
||||
/**
|
||||
* Does audio stretching and produces the time-stretched samples.
|
||||
* Timer calculations use sample_delay to determine how much of a margin we have.
|
||||
* @param sample_delay How many samples are buffered downstream of this module and haven't been
|
||||
* played yet.
|
||||
* @return Samples to play in interleaved stereo PCM16 format.
|
||||
*/
|
||||
std::vector<s16> Process(std::size_t sample_delay);
|
||||
|
||||
private:
|
||||
struct Impl;
|
||||
std::unique_ptr<Impl> impl;
|
||||
|
||||
/// INTERNAL: ratio = wallclock time / emulated time
|
||||
double CalculateCurrentRatio();
|
||||
/// INTERNAL: If we have too many or too few samples downstream, nudge ratio in the appropriate
|
||||
/// direction.
|
||||
double CorrectForUnderAndOverflow(double ratio, std::size_t sample_delay) const;
|
||||
/// INTERNAL: Gets the time-stretched samples from SoundTouch.
|
||||
std::vector<s16> GetSamples();
|
||||
unsigned int sample_rate;
|
||||
std::unique_ptr<soundtouch::SoundTouch> sound_touch;
|
||||
double stretch_ratio = 1.0;
|
||||
};
|
||||
|
||||
} // namespace AudioCore
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue