21 files changed, 462 insertions, 49 deletions

diff --git a/.gitmodules b/.gitmodules
index 4f4e8690b..e73ca99e3 100644
--- a/.gitmodules
+++ b/.gitmodules
@@ -31,3 +31,6 @@
 [submodule "opus"]
     path = externals/opus
     url = https://github.com/ogniK5377/opus.git
+[submodule "soundtouch"]
+        path = externals/soundtouch
+        url = https://github.com/citra-emu/ext-soundtouch.git
diff --git a/externals/CMakeLists.txt b/externals/CMakeLists.txt
index 3d8e10c2b..53dcf1f1a 100644
--- a/externals/CMakeLists.txt
+++ b/externals/CMakeLists.txt
@@ -50,6 +50,9 @@ add_subdirectory(open_source_archives EXCLUDE_FROM_ALL)
 add_library(unicorn-headers INTERFACE)
 target_include_directories(unicorn-headers INTERFACE ./unicorn/include)
 
+# SoundTouch
+add_subdirectory(soundtouch)
+
 # Xbyak
 if (ARCHITECTURE_x86_64)
     # Defined before "dynarmic" above
diff --git a/externals/soundtouch b/externals/soundtouch
new file mode 160000
+Subproject 060181eaf273180d3a7e87349895bd0cb6ccbf4
diff --git a/src/audio_core/CMakeLists.txt b/src/audio_core/CMakeLists.txt
index 82e4850f7..c381dbe1d 100644
--- a/src/audio_core/CMakeLists.txt
+++ b/src/audio_core/CMakeLists.txt
@@ -17,6 +17,8 @@ add_library(audio_core STATIC
     sink_stream.h
     stream.cpp
     stream.h
+    time_stretch.cpp
+    time_stretch.h
 
     $<$<BOOL:${ENABLE_CUBEB}>:cubeb_sink.cpp cubeb_sink.h>
 )
@@ -24,6 +26,7 @@ add_library(audio_core STATIC
 create_target_directory_groups(audio_core)
 
 target_link_libraries(audio_core PUBLIC common core)
+target_link_libraries(audio_core PRIVATE SoundTouch)
 
 if(ENABLE_CUBEB)
     target_link_libraries(audio_core PRIVATE cubeb)
diff --git a/src/audio_core/cubeb_sink.cpp b/src/audio_core/cubeb_sink.cpp
index 5a1177d0c..79155a7a0 100644
--- a/src/audio_core/cubeb_sink.cpp
+++ b/src/audio_core/cubeb_sink.cpp
@@ -3,27 +3,23 @@
 // Refer to the license.txt file included.
 
 #include <algorithm>
+#include <atomic>
 #include <cstring>
-#include <mutex>
-
 #include "audio_core/cubeb_sink.h"
 #include "audio_core/stream.h"
+#include "audio_core/time_stretch.h"
 #include "common/logging/log.h"
+#include "common/ring_buffer.h"
+#include "core/settings.h"
 
 namespace AudioCore {
 
-class SinkStreamImpl final : public SinkStream {
+class CubebSinkStream final : public SinkStream {
 public:
-    SinkStreamImpl(cubeb* ctx, u32 sample_rate, u32 num_channels_, cubeb_devid output_device,
-                   const std::string& name)
-        : ctx{ctx}, num_channels{num_channels_} {
-
-        if (num_channels == 6) {
-            // 6-channel audio does not seem to work with cubeb + SDL, so we downsample this to 2
-            // channel for now
-            is_6_channel = true;
-            num_channels = 2;
-        }
+    CubebSinkStream(cubeb* ctx, u32 sample_rate, u32 num_channels_, cubeb_devid output_device,
+                    const std::string& name)
+        : ctx{ctx}, num_channels{std::min(num_channels_, 2u)}, time_stretch{sample_rate,
+                                                                            num_channels} {
 
         cubeb_stream_params params{};
         params.rate = sample_rate;
@@ -38,7 +34,7 @@ public:
 
         if (cubeb_stream_init(ctx, &stream_backend, name.c_str(), nullptr, nullptr, output_device,
                               &params, std::max(512u, minimum_latency),
-                              &SinkStreamImpl::DataCallback, &SinkStreamImpl::StateCallback,
+                              &CubebSinkStream::DataCallback, &CubebSinkStream::StateCallback,
                               this) != CUBEB_OK) {
             LOG_CRITICAL(Audio_Sink, "Error initializing cubeb stream");
             return;
@@ -50,7 +46,7 @@ public:
         }
     }
 
-    ~SinkStreamImpl() {
+    ~CubebSinkStream() {
         if (!ctx) {
             return;
         }
@@ -62,27 +58,32 @@ public:
         cubeb_stream_destroy(stream_backend);
     }
 
-    void EnqueueSamples(u32 num_channels, const std::vector<s16>& samples) override {
-        if (!ctx) {
+    void EnqueueSamples(u32 source_num_channels, const std::vector<s16>& samples) override {
+        if (source_num_channels > num_channels) {
+            // Downsample 6 channels to 2
+            std::vector<s16> buf;
+            buf.reserve(samples.size() * num_channels / source_num_channels);
+            for (size_t i = 0; i < samples.size(); i += source_num_channels) {
+                for (size_t ch = 0; ch < num_channels; ch++) {
+                    buf.push_back(samples[i + ch]);
+                }
+            }
+            queue.Push(buf);
             return;
         }
 
-        std::lock_guard lock{queue_mutex};
+        queue.Push(samples);
+    }
 
-        queue.reserve(queue.size() + samples.size() * GetNumChannels());
+    size_t SamplesInQueue(u32 num_channels) const override {
+        if (!ctx)
+            return 0;
 
-        if (is_6_channel) {
-            // Downsample 6 channels to 2
-            const size_t sample_count_copy_size = samples.size() * 2;
-            queue.reserve(sample_count_copy_size);
-            for (size_t i = 0; i < samples.size(); i += num_channels) {
-                queue.push_back(samples[i]);
-                queue.push_back(samples[i + 1]);
-            }
-        } else {
-            // Copy as-is
-            std::copy(samples.begin(), samples.end(), std::back_inserter(queue));
-        }
+        return queue.Size() / num_channels;
+    }
+
+    void Flush() override {
+        should_flush = true;
     }
 
     u32 GetNumChannels() const {
@@ -95,10 +96,11 @@ private:
     cubeb* ctx{};
     cubeb_stream* stream_backend{};
     u32 num_channels{};
-    bool is_6_channel{};
 
-    std::mutex queue_mutex;
-    std::vector<s16> queue;
+    Common::RingBuffer<s16, 0x10000> queue;
+    std::array<s16, 2> last_frame;
+    std::atomic<bool> should_flush{};
+    TimeStretcher time_stretch;
 
     static long DataCallback(cubeb_stream* stream, void* user_data, const void* input_buffer,
                              void* output_buffer, long num_frames);
@@ -144,38 +146,52 @@ CubebSink::~CubebSink() {
 SinkStream& CubebSink::AcquireSinkStream(u32 sample_rate, u32 num_channels,
                                          const std::string& name) {
     sink_streams.push_back(
-        std::make_unique<SinkStreamImpl>(ctx, sample_rate, num_channels, output_device, name));
+        std::make_unique<CubebSinkStream>(ctx, sample_rate, num_channels, output_device, name));
     return *sink_streams.back();
 }
 
-long SinkStreamImpl::DataCallback(cubeb_stream* stream, void* user_data, const void* input_buffer,
-                                  void* output_buffer, long num_frames) {
-    SinkStreamImpl* impl = static_cast<SinkStreamImpl*>(user_data);
+long CubebSinkStream::DataCallback(cubeb_stream* stream, void* user_data, const void* input_buffer,
+                                   void* output_buffer, long num_frames) {
+    CubebSinkStream* impl = static_cast<CubebSinkStream*>(user_data);
     u8* buffer = reinterpret_cast<u8*>(output_buffer);
 
     if (!impl) {
         return {};
     }
 
-    std::lock_guard lock{impl->queue_mutex};
+    const size_t num_channels = impl->GetNumChannels();
+    const size_t samples_to_write = num_channels * num_frames;
+    size_t samples_written;
+
+    if (Settings::values.enable_audio_stretching) {
+        const std::vector<s16> in{impl->queue.Pop()};
+        const size_t num_in{in.size() / num_channels};
+        s16* const out{reinterpret_cast<s16*>(buffer)};
+        const size_t out_frames = impl->time_stretch.Process(in.data(), num_in, out, num_frames);
+        samples_written = out_frames * num_channels;
 
-    const size_t frames_to_write{
-        std::min(impl->queue.size() / impl->GetNumChannels(), static_cast<size_t>(num_frames))};
+        if (impl->should_flush) {
+            impl->time_stretch.Flush();
+            impl->should_flush = false;
+        }
+    } else {
+        samples_written = impl->queue.Pop(buffer, samples_to_write);
+    }
 
-    memcpy(buffer, impl->queue.data(), frames_to_write * sizeof(s16) * impl->GetNumChannels());
-    impl->queue.erase(impl->queue.begin(),
-                      impl->queue.begin() + frames_to_write * impl->GetNumChannels());
+    if (samples_written >= num_channels) {
+        std::memcpy(&impl->last_frame[0], buffer + (samples_written - num_channels) * sizeof(s16),
+                    num_channels * sizeof(s16));
+    }
 
-    if (frames_to_write < num_frames) {
-        // Fill the rest of the frames with silence
-        memset(buffer + frames_to_write * sizeof(s16) * impl->GetNumChannels(), 0,
-               (num_frames - frames_to_write) * sizeof(s16) * impl->GetNumChannels());
+    // Fill the rest of the frames with last_frame
+    for (size_t i = samples_written; i < samples_to_write; i += num_channels) {
+        std::memcpy(buffer + i * sizeof(s16), &impl->last_frame[0], num_channels * sizeof(s16));
     }
 
     return num_frames;
 }
 
-void SinkStreamImpl::StateCallback(cubeb_stream* stream, void* user_data, cubeb_state state) {}
+void CubebSinkStream::StateCallback(cubeb_stream* stream, void* user_data, cubeb_state state) {}
 
 std::vector<std::string> ListCubebSinkDevices() {
     std::vector<std::string> device_list;
diff --git a/src/audio_core/null_sink.h b/src/audio_core/null_sink.h
index f235d93e5..2ed0c83b6 100644
--- a/src/audio_core/null_sink.h
+++ b/src/audio_core/null_sink.h
@@ -21,6 +21,12 @@ public:
 private:
     struct NullSinkStreamImpl final : SinkStream {
         void EnqueueSamples(u32 /*num_channels*/, const std::vector<s16>& /*samples*/) override {}
+
+        size_t SamplesInQueue(u32 /*num_channels*/) const override {
+            return 0;
+        }
+
+        void Flush() override {}
     } null_sink_stream;
 };
 
diff --git a/src/audio_core/sink_stream.h b/src/audio_core/sink_stream.h
index 41b6736d8..4309ad094 100644
--- a/src/audio_core/sink_stream.h
+++ b/src/audio_core/sink_stream.h
@@ -25,6 +25,10 @@ public:
      * @param samples Samples in interleaved stereo PCM16 format.
      */
     virtual void EnqueueSamples(u32 num_channels, const std::vector<s16>& samples) = 0;
+
+    virtual std::size_t SamplesInQueue(u32 num_channels) const = 0;
+
+    virtual void Flush() = 0;
 };
 
 using SinkStreamPtr = std::unique_ptr<SinkStream>;
diff --git a/src/audio_core/stream.cpp b/src/audio_core/stream.cpp
index dbae75d8c..84dcdd98d 100644
--- a/src/audio_core/stream.cpp
+++ b/src/audio_core/stream.cpp
@@ -73,6 +73,7 @@ static void VolumeAdjustSamples(std::vector<s16>& samples) {
 void Stream::PlayNextBuffer() {
     if (!IsPlaying()) {
         // Ensure we are in playing state before playing the next buffer
+        sink_stream.Flush();
         return;
     }
 
@@ -83,6 +84,7 @@ void Stream::PlayNextBuffer() {
 
     if (queued_buffers.empty()) {
         // No queued buffers - we are effectively paused
+        sink_stream.Flush();
         return;
     }
 
@@ -90,6 +92,7 @@ void Stream::PlayNextBuffer() {
     queued_buffers.pop();
 
     VolumeAdjustSamples(active_buffer->Samples());
+
     sink_stream.EnqueueSamples(GetNumChannels(), active_buffer->GetSamples());
 
     CoreTiming::ScheduleEventThreadsafe(GetBufferReleaseCycles(*active_buffer), release_event, {});
diff --git a/src/audio_core/time_stretch.cpp b/src/audio_core/time_stretch.cpp
new file mode 100644
index 000000000..da094c46b
--- /dev/null
+++ b/src/audio_core/time_stretch.cpp
@@ -0,0 +1,68 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <cmath>
+#include <cstddef>
+#include "audio_core/time_stretch.h"
+#include "common/logging/log.h"
+
+namespace AudioCore {
+
+TimeStretcher::TimeStretcher(u32 sample_rate, u32 channel_count)
+    : m_sample_rate(sample_rate), m_channel_count(channel_count) {
+    m_sound_touch.setChannels(channel_count);
+    m_sound_touch.setSampleRate(sample_rate);
+    m_sound_touch.setPitch(1.0);
+    m_sound_touch.setTempo(1.0);
+}
+
+void TimeStretcher::Clear() {
+    m_sound_touch.clear();
+}
+
+void TimeStretcher::Flush() {
+    m_sound_touch.flush();
+}
+
+size_t TimeStretcher::Process(const s16* in, size_t num_in, s16* out, size_t 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 = 1.0; // seconds
+    const double max_backlog = m_sample_rate * max_latency;
+    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 = 0.05; // 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 = 2.0; // 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 5% 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, 0.05);
+    m_sound_touch.setTempo(m_stretch_ratio);
+
+    LOG_DEBUG(Audio, "{:5}/{:5} ratio:{:0.6f} backlog:{:0.6f}", num_in, num_out, m_stretch_ratio,
+              backlog_fullness);
+
+    m_sound_touch.putSamples(in, num_in);
+    return m_sound_touch.receiveSamples(out, num_out);
+}
+
+} // namespace AudioCore
diff --git a/src/audio_core/time_stretch.h b/src/audio_core/time_stretch.h
new file mode 100644
index 000000000..7e39e695e
--- /dev/null
+++ b/src/audio_core/time_stretch.h
@@ -0,0 +1,36 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <cstddef>
+#include <SoundTouch.h>
+#include "common/common_types.h"
+
+namespace AudioCore {
+
+class TimeStretcher {
+public:
+    TimeStretcher(u32 sample_rate, u32 channel_count);
+
+    /// @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`
+    size_t Process(const s16* in, size_t num_in, s16* out, size_t num_out);
+
+    void Clear();
+
+    void Flush();
+
+private:
+    u32 m_sample_rate;
+    u32 m_channel_count;
+    soundtouch::SoundTouch m_sound_touch;
+    double m_stretch_ratio = 1.0;
+};
+
+} // namespace AudioCore
diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
index f41946cc6..6a3f1fe08 100644
--- a/src/common/CMakeLists.txt
+++ b/src/common/CMakeLists.txt
@@ -71,6 +71,7 @@ add_library(common STATIC
     param_package.cpp
     param_package.h
     quaternion.h
+    ring_buffer.h
     scm_rev.cpp
     scm_rev.h
     scope_exit.h
diff --git a/src/common/ring_buffer.h b/src/common/ring_buffer.h
new file mode 100644
index 000000000..30d934a38
--- /dev/null
+++ b/src/common/ring_buffer.h
@@ -0,0 +1,111 @@
+// Copyright 2018 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <algorithm>
+#include <array>
+#include <atomic>
+#include <cstddef>
+#include <cstring>
+#include <type_traits>
+#include <vector>
+#include "common/common_types.h"
+
+namespace Common {
+
+/// SPSC ring buffer
+/// @tparam T            Element type
+/// @tparam capacity     Number of slots in ring buffer
+/// @tparam granularity  Slot size in terms of number of elements
+template <typename T, size_t capacity, size_t granularity = 1>
+class RingBuffer {
+    /// A "slot" is made of `granularity` elements of `T`.
+    static constexpr size_t slot_size = granularity * sizeof(T);
+    // T must be safely memcpy-able and have a trivial default constructor.
+    static_assert(std::is_trivial_v<T>);
+    // Ensure capacity is sensible.
+    static_assert(capacity < std::numeric_limits<size_t>::max() / 2 / granularity);
+    static_assert((capacity & (capacity - 1)) == 0, "capacity must be a power of two");
+    // Ensure lock-free.
+    static_assert(std::atomic<size_t>::is_always_lock_free);
+
+public:
+    /// Pushes slots into the ring buffer
+    /// @param new_slots   Pointer to the slots to push
+    /// @param slot_count  Number of slots to push
+    /// @returns The number of slots actually pushed
+    size_t Push(const void* new_slots, size_t slot_count) {
+        const size_t write_index = m_write_index.load();
+        const size_t slots_free = capacity + m_read_index.load() - write_index;
+        const size_t push_count = std::min(slot_count, slots_free);
+
+        const size_t pos = write_index % capacity;
+        const size_t first_copy = std::min(capacity - pos, push_count);
+        const size_t second_copy = push_count - first_copy;
+
+        const char* in = static_cast<const char*>(new_slots);
+        std::memcpy(m_data.data() + pos * granularity, in, first_copy * slot_size);
+        in += first_copy * slot_size;
+        std::memcpy(m_data.data(), in, second_copy * slot_size);
+
+        m_write_index.store(write_index + push_count);
+
+        return push_count;
+    }
+
+    size_t Push(const std::vector<T>& input) {
+        return Push(input.data(), input.size());
+    }
+
+    /// Pops slots from the ring buffer
+    /// @param output     Where to store the popped slots
+    /// @param max_slots  Maximum number of slots to pop
+    /// @returns The number of slots actually popped
+    size_t Pop(void* output, size_t max_slots = ~size_t(0)) {
+        const size_t read_index = m_read_index.load();
+        const size_t slots_filled = m_write_index.load() - read_index;
+        const size_t pop_count = std::min(slots_filled, max_slots);
+
+        const size_t pos = read_index % capacity;
+        const size_t first_copy = std::min(capacity - pos, pop_count);
+        const size_t second_copy = pop_count - first_copy;
+
+        char* out = static_cast<char*>(output);
+        std::memcpy(out, m_data.data() + pos * granularity, first_copy * slot_size);
+        out += first_copy * slot_size;
+        std::memcpy(out, m_data.data(), second_copy * slot_size);
+
+        m_read_index.store(read_index + pop_count);
+
+        return pop_count;
+    }
+
+    std::vector<T> Pop(size_t max_slots = ~size_t(0)) {
+        std::vector<T> out(std::min(max_slots, capacity) * granularity);
+        const size_t count = Pop(out.data(), out.size() / granularity);
+        out.resize(count * granularity);
+        return out;
+    }
+
+    /// @returns Number of slots used
+    size_t Size() const {
+        return m_write_index.load() - m_read_index.load();
+    }
+
+    /// @returns Maximum size of ring buffer
+    constexpr size_t Capacity() const {
+        return capacity;
+    }
+
+private:
+    // It is important to align the below variables for performance reasons:
+    // Having them on the same cache-line would result in false-sharing between them.
+    alignas(128) std::atomic<size_t> m_read_index{0};
+    alignas(128) std::atomic<size_t> m_write_index{0};
+
+    std::array<T, granularity * capacity> m_data;
+};
+
+} // namespace Common
diff --git a/src/core/settings.h b/src/core/settings.h
index 08a16ef2c..0318d019c 100644
--- a/src/core/settings.h
+++ b/src/core/settings.h
@@ -148,6 +148,7 @@ struct Values {
 
     // Audio
     std::string sink_id;
+    bool enable_audio_stretching;
     std::string audio_device_id;
     float volume;
 
diff --git a/src/core/telemetry_session.cpp b/src/core/telemetry_session.cpp
index 3730e85b8..b0df154ca 100644
--- a/src/core/telemetry_session.cpp
+++ b/src/core/telemetry_session.cpp
@@ -120,6 +120,9 @@ TelemetrySession::TelemetrySession() {
     Telemetry::AppendOSInfo(field_collection);
 
     // Log user configuration information
+    AddField(Telemetry::FieldType::UserConfig, "Audio_SinkId", Settings::values.sink_id);
+    AddField(Telemetry::FieldType::UserConfig, "Audio_EnableAudioStretching",
+             Settings::values.enable_audio_stretching);
     AddField(Telemetry::FieldType::UserConfig, "Core_UseCpuJit", Settings::values.use_cpu_jit);
     AddField(Telemetry::FieldType::UserConfig, "Core_UseMultiCore",
              Settings::values.use_multi_core);
diff --git a/src/tests/CMakeLists.txt b/src/tests/CMakeLists.txt
index 4d74bb395..4e75a72ec 100644
--- a/src/tests/CMakeLists.txt
+++ b/src/tests/CMakeLists.txt
@@ -1,5 +1,6 @@
 add_executable(tests
     common/param_package.cpp
+    common/ring_buffer.cpp
     core/arm/arm_test_common.cpp
     core/arm/arm_test_common.h
     core/core_timing.cpp
diff --git a/src/tests/common/ring_buffer.cpp b/src/tests/common/ring_buffer.cpp
new file mode 100644
index 000000000..f3fe57839
--- /dev/null
+++ b/src/tests/common/ring_buffer.cpp
@@ -0,0 +1,130 @@
+// Copyright 2018 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <array>
+#include <cstddef>
+#include <numeric>
+#include <thread>
+#include <vector>
+#include <catch2/catch.hpp>
+#include "common/ring_buffer.h"
+
+namespace Common {
+
+TEST_CASE("RingBuffer: Basic Tests", "[common]") {
+    RingBuffer<char, 4, 1> buf;
+
+    // Pushing values into a ring buffer with space should succeed.
+    for (size_t i = 0; i < 4; i++) {
+        const char elem = static_cast<char>(i);
+        const size_t count = buf.Push(&elem, 1);
+        REQUIRE(count == 1);
+    }
+
+    REQUIRE(buf.Size() == 4);
+
+    // Pushing values into a full ring buffer should fail.
+    {
+        const char elem = static_cast<char>(42);
+        const size_t count = buf.Push(&elem, 1);
+        REQUIRE(count == 0);
+    }
+
+    REQUIRE(buf.Size() == 4);
+
+    // Popping multiple values from a ring buffer with values should succeed.
+    {
+        const std::vector<char> popped = buf.Pop(2);
+        REQUIRE(popped.size() == 2);
+        REQUIRE(popped[0] == 0);
+        REQUIRE(popped[1] == 1);
+    }
+
+    REQUIRE(buf.Size() == 2);
+
+    // Popping a single value from a ring buffer with values should succeed.
+    {
+        const std::vector<char> popped = buf.Pop(1);
+        REQUIRE(popped.size() == 1);
+        REQUIRE(popped[0] == 2);
+    }
+
+    REQUIRE(buf.Size() == 1);
+
+    // Pushing more values than space available should partially suceed.
+    {
+        std::vector<char> to_push(6);
+        std::iota(to_push.begin(), to_push.end(), 88);
+        const size_t count = buf.Push(to_push);
+        REQUIRE(count == 3);
+    }
+
+    REQUIRE(buf.Size() == 4);
+
+    // Doing an unlimited pop should pop all values.
+    {
+        const std::vector<char> popped = buf.Pop();
+        REQUIRE(popped.size() == 4);
+        REQUIRE(popped[0] == 3);
+        REQUIRE(popped[1] == 88);
+        REQUIRE(popped[2] == 89);
+        REQUIRE(popped[3] == 90);
+    }
+
+    REQUIRE(buf.Size() == 0);
+}
+
+TEST_CASE("RingBuffer: Threaded Test", "[common]") {
+    RingBuffer<char, 4, 2> buf;
+    const char seed = 42;
+    const size_t count = 1000000;
+    size_t full = 0;
+    size_t empty = 0;
+
+    const auto next_value = [](std::array<char, 2>& value) {
+        value[0] += 1;
+        value[1] += 2;
+    };
+
+    std::thread producer{[&] {
+        std::array<char, 2> value = {seed, seed};
+        size_t i = 0;
+        while (i < count) {
+            if (const size_t c = buf.Push(&value[0], 1); c > 0) {
+                REQUIRE(c == 1);
+                i++;
+                next_value(value);
+            } else {
+                full++;
+                std::this_thread::yield();
+            }
+        }
+    }};
+
+    std::thread consumer{[&] {
+        std::array<char, 2> value = {seed, seed};
+        size_t i = 0;
+        while (i < count) {
+            if (const std::vector<char> v = buf.Pop(1); v.size() > 0) {
+                REQUIRE(v.size() == 2);
+                REQUIRE(v[0] == value[0]);
+                REQUIRE(v[1] == value[1]);
+                i++;
+                next_value(value);
+            } else {
+                empty++;
+                std::this_thread::yield();
+            }
+        }
+    }};
+
+    producer.join();
+    consumer.join();
+
+    REQUIRE(buf.Size() == 0);
+    printf("RingBuffer: Threaded Test: full: %zu, empty: %zu\n", full, empty);
+}
+
+} // namespace Common
diff --git a/src/yuzu/configuration/config.cpp b/src/yuzu/configuration/config.cpp
index c43e79e78..d229225b4 100644
--- a/src/yuzu/configuration/config.cpp
+++ b/src/yuzu/configuration/config.cpp
@@ -95,6 +95,8 @@ void Config::ReadValues() {
 
     qt_config->beginGroup("Audio");
     Settings::values.sink_id = qt_config->value("output_engine", "auto").toString().toStdString();
+    Settings::values.enable_audio_stretching =
+        qt_config->value("enable_audio_stretching", true).toBool();
     Settings::values.audio_device_id =
         qt_config->value("output_device", "auto").toString().toStdString();
     Settings::values.volume = qt_config->value("volume", 1).toFloat();
@@ -230,6 +232,7 @@ void Config::SaveValues() {
 
     qt_config->beginGroup("Audio");
     qt_config->setValue("output_engine", QString::fromStdString(Settings::values.sink_id));
+    qt_config->setValue("enable_audio_stretching", Settings::values.enable_audio_stretching);
     qt_config->setValue("output_device", QString::fromStdString(Settings::values.audio_device_id));
     qt_config->setValue("volume", Settings::values.volume);
     qt_config->endGroup();
diff --git a/src/yuzu/configuration/configure_audio.cpp b/src/yuzu/configuration/configure_audio.cpp
index fbb813f6c..6ea59f2a3 100644
--- a/src/yuzu/configuration/configure_audio.cpp
+++ b/src/yuzu/configuration/configure_audio.cpp
@@ -46,6 +46,8 @@ void ConfigureAudio::setConfiguration() {
     }
     ui->output_sink_combo_box->setCurrentIndex(new_sink_index);
 
+    ui->toggle_audio_stretching->setChecked(Settings::values.enable_audio_stretching);
+
     // The device list cannot be pre-populated (nor listed) until the output sink is known.
     updateAudioDevices(new_sink_index);
 
@@ -67,6 +69,7 @@ void ConfigureAudio::applyConfiguration() {
     Settings::values.sink_id =
         ui->output_sink_combo_box->itemText(ui->output_sink_combo_box->currentIndex())
             .toStdString();
+    Settings::values.enable_audio_stretching = ui->toggle_audio_stretching->isChecked();
     Settings::values.audio_device_id =
         ui->audio_device_combo_box->itemText(ui->audio_device_combo_box->currentIndex())
             .toStdString();
diff --git a/src/yuzu/configuration/configure_audio.ui b/src/yuzu/configuration/configure_audio.ui
index ef67890dc..a29a0e265 100644
--- a/src/yuzu/configuration/configure_audio.ui
+++ b/src/yuzu/configuration/configure_audio.ui
@@ -31,6 +31,16 @@
         </item>
        </layout>
       </item>
+       <item>
+         <widget class="QCheckBox" name="toggle_audio_stretching">
+           <property name="toolTip">
+             <string>This post-processing effect adjusts audio speed to match emulation speed and helps prevent audio stutter. This however increases audio latency.</string>
+           </property>
+           <property name="text">
+             <string>Enable audio stretching</string>
+           </property>
+         </widget>
+       </item>
       <item>
        <layout class="QHBoxLayout">
         <item>
diff --git a/src/yuzu_cmd/config.cpp b/src/yuzu_cmd/config.cpp
index f00b5a66b..991abda2e 100644
--- a/src/yuzu_cmd/config.cpp
+++ b/src/yuzu_cmd/config.cpp
@@ -108,6 +108,8 @@ void Config::ReadValues() {
 
     // Audio
     Settings::values.sink_id = sdl2_config->Get("Audio", "output_engine", "auto");
+    Settings::values.enable_audio_stretching =
+        sdl2_config->GetBoolean("Audio", "enable_audio_stretching", true);
     Settings::values.audio_device_id = sdl2_config->Get("Audio", "output_device", "auto");
     Settings::values.volume = sdl2_config->GetReal("Audio", "volume", 1);
 
diff --git a/src/yuzu_cmd/default_ini.h b/src/yuzu_cmd/default_ini.h
index 6ed9e7962..002a4ec15 100644
--- a/src/yuzu_cmd/default_ini.h
+++ b/src/yuzu_cmd/default_ini.h
@@ -150,6 +150,12 @@ swap_screen =
 # auto (default): Auto-select, null: No audio output, cubeb: Cubeb audio engine (if available)
 output_engine =
 
+# Whether or not to enable the audio-stretching post-processing effect.
+# This effect adjusts audio speed to match emulation speed and helps prevent audio stutter,
+# at the cost of increasing audio latency.
+# 0: No, 1 (default): Yes
+enable_audio_stretching =
+
 # Which audio device to use.
 # auto (default): Auto-select
 output_device =