9 files changed, 1934 insertions, 0 deletions

diff --git a/src/audio_core/sink/cubeb_sink.cpp b/src/audio_core/sink/cubeb_sink.cpp
new file mode 100644
index 000000000..a4e28de6d
--- /dev/null
+++ b/src/audio_core/sink/cubeb_sink.cpp
@@ -0,0 +1,651 @@
+// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <algorithm>
+#include <atomic>
+#include <span>
+
+#include "audio_core/audio_core.h"
+#include "audio_core/audio_event.h"
+#include "audio_core/audio_manager.h"
+#include "audio_core/sink/cubeb_sink.h"
+#include "audio_core/sink/sink_stream.h"
+#include "common/assert.h"
+#include "common/fixed_point.h"
+#include "common/logging/log.h"
+#include "common/reader_writer_queue.h"
+#include "common/ring_buffer.h"
+#include "common/settings.h"
+#include "core/core.h"
+
+#ifdef _WIN32
+#include <objbase.h>
+#undef CreateEvent
+#endif
+
+namespace AudioCore::Sink {
+/**
+ * Cubeb sink stream, responsible for sinking samples to hardware.
+ */
+class CubebSinkStream final : public SinkStream {
+public:
+    /**
+     * Create a new sink stream.
+     *
+     * @param ctx_             - Cubeb context to create this stream with.
+     * @param device_channels_ - Number of channels supported by the hardware.
+     * @param system_channels_ - Number of channels the audio systems expect.
+     * @param output_device    - Cubeb output device id.
+     * @param input_device     - Cubeb input device id.
+     * @param name_            - Name of this stream.
+     * @param type_            - Type of this stream.
+     * @param system_          - Core system.
+     * @param event            - Event used only for audio renderer, signalled on buffer consume.
+     */
+    CubebSinkStream(cubeb* ctx_, const u32 device_channels_, const u32 system_channels_,
+                    cubeb_devid output_device, cubeb_devid input_device, const std::string& name_,
+                    const StreamType type_, Core::System& system_)
+        : ctx{ctx_}, type{type_}, system{system_} {
+#ifdef _WIN32
+        CoInitializeEx(nullptr, COINIT_MULTITHREADED);
+#endif
+        name = name_;
+        device_channels = device_channels_;
+        system_channels = system_channels_;
+
+        cubeb_stream_params params{};
+        params.rate = TargetSampleRate;
+        params.channels = device_channels;
+        params.format = CUBEB_SAMPLE_S16LE;
+        params.prefs = CUBEB_STREAM_PREF_NONE;
+        switch (params.channels) {
+        case 1:
+            params.layout = CUBEB_LAYOUT_MONO;
+            break;
+        case 2:
+            params.layout = CUBEB_LAYOUT_STEREO;
+            break;
+        case 6:
+            params.layout = CUBEB_LAYOUT_3F2_LFE;
+            break;
+        }
+
+        u32 minimum_latency{0};
+        const auto latency_error = cubeb_get_min_latency(ctx, &params, &minimum_latency);
+        if (latency_error != CUBEB_OK) {
+            LOG_CRITICAL(Audio_Sink, "Error getting minimum latency, error: {}", latency_error);
+            minimum_latency = 256U;
+        }
+
+        minimum_latency = std::max(minimum_latency, 256u);
+
+        playing_buffer.consumed = true;
+
+        LOG_DEBUG(Service_Audio,
+                  "Opening cubeb stream {} type {} with: rate {} channels {} (system channels {}) "
+                  "latency {}",
+                  name, type, params.rate, params.channels, system_channels, minimum_latency);
+
+        auto init_error{0};
+        if (type == StreamType::In) {
+            init_error = cubeb_stream_init(ctx, &stream_backend, name.c_str(), input_device,
+                                           &params, output_device, nullptr, minimum_latency,
+                                           &CubebSinkStream::DataCallback,
+                                           &CubebSinkStream::StateCallback, this);
+        } else {
+            init_error = cubeb_stream_init(ctx, &stream_backend, name.c_str(), input_device,
+                                           nullptr, output_device, &params, minimum_latency,
+                                           &CubebSinkStream::DataCallback,
+                                           &CubebSinkStream::StateCallback, this);
+        }
+
+        if (init_error != CUBEB_OK) {
+            LOG_CRITICAL(Audio_Sink, "Error initializing cubeb stream, error: {}", init_error);
+            return;
+        }
+    }
+
+    /**
+     * Destroy the sink stream.
+     */
+    ~CubebSinkStream() override {
+        LOG_DEBUG(Service_Audio, "Destructing cubeb stream {}", name);
+
+        if (!ctx) {
+            return;
+        }
+
+        Finalize();
+
+#ifdef _WIN32
+        CoUninitialize();
+#endif
+    }
+
+    /**
+     * Finalize the sink stream.
+     */
+    void Finalize() override {
+        Stop();
+        cubeb_stream_destroy(stream_backend);
+    }
+
+    /**
+     * Start the sink stream.
+     *
+     * @param resume - Set to true if this is resuming the stream a previously-active stream.
+     *                 Default false.
+     */
+    void Start(const bool resume = false) override {
+        if (!ctx) {
+            return;
+        }
+
+        if (resume && was_playing) {
+            if (cubeb_stream_start(stream_backend) != CUBEB_OK) {
+                LOG_CRITICAL(Audio_Sink, "Error starting cubeb stream");
+            }
+            paused = false;
+        } else if (!resume) {
+            if (cubeb_stream_start(stream_backend) != CUBEB_OK) {
+                LOG_CRITICAL(Audio_Sink, "Error starting cubeb stream");
+            }
+            paused = false;
+        }
+    }
+
+    /**
+     * Stop the sink stream.
+     */
+    void Stop() override {
+        if (!ctx) {
+            return;
+        }
+
+        if (cubeb_stream_stop(stream_backend) != CUBEB_OK) {
+            LOG_CRITICAL(Audio_Sink, "Error stopping cubeb stream");
+        }
+
+        was_playing.store(!paused);
+        paused = true;
+    }
+
+    /**
+     * Append a new buffer and its samples to a waiting queue to play.
+     *
+     * @param buffer  - Audio buffer information to be queued.
+     * @param samples - The s16 samples to be queue for playback.
+     */
+    void AppendBuffer(::AudioCore::Sink::SinkBuffer& buffer, std::vector<s16>& samples) override {
+        if (type == StreamType::In) {
+            queue.enqueue(buffer);
+            queued_buffers++;
+        } else {
+            constexpr s32 min{std::numeric_limits<s16>::min()};
+            constexpr s32 max{std::numeric_limits<s16>::max()};
+
+            auto yuzu_volume{Settings::Volume()};
+            auto volume{system_volume * device_volume * yuzu_volume};
+
+            if (system_channels == 6 && device_channels == 2) {
+                // We're given 6 channels, but our device only outputs 2, so downmix.
+                constexpr std::array<f32, 4> down_mix_coeff{1.0f, 0.707f, 0.251f, 0.707f};
+
+                for (u32 read_index = 0, write_index = 0; read_index < samples.size();
+                     read_index += system_channels, write_index += device_channels) {
+                    const auto left_sample{
+                        ((Common::FixedPoint<49, 15>(
+                              samples[read_index + static_cast<u32>(Channels::FrontLeft)]) *
+                              down_mix_coeff[0] +
+                          samples[read_index + static_cast<u32>(Channels::Center)] *
+                              down_mix_coeff[1] +
+                          samples[read_index + static_cast<u32>(Channels::LFE)] *
+                              down_mix_coeff[2] +
+                          samples[read_index + static_cast<u32>(Channels::BackLeft)] *
+                              down_mix_coeff[3]) *
+                         volume)
+                            .to_int()};
+
+                    const auto right_sample{
+                        ((Common::FixedPoint<49, 15>(
+                              samples[read_index + static_cast<u32>(Channels::FrontRight)]) *
+                              down_mix_coeff[0] +
+                          samples[read_index + static_cast<u32>(Channels::Center)] *
+                              down_mix_coeff[1] +
+                          samples[read_index + static_cast<u32>(Channels::LFE)] *
+                              down_mix_coeff[2] +
+                          samples[read_index + static_cast<u32>(Channels::BackRight)] *
+                              down_mix_coeff[3]) *
+                         volume)
+                            .to_int()};
+
+                    samples[write_index + static_cast<u32>(Channels::FrontLeft)] =
+                        static_cast<s16>(std::clamp(left_sample, min, max));
+                    samples[write_index + static_cast<u32>(Channels::FrontRight)] =
+                        static_cast<s16>(std::clamp(right_sample, min, max));
+                }
+
+                samples.resize(samples.size() / system_channels * device_channels);
+
+            } else if (system_channels == 2 && device_channels == 6) {
+                // We need moar samples! Not all games will provide 6 channel audio.
+                // TODO: Implement some upmixing here. Currently just passthrough, with other
+                // channels left as silence.
+                std::vector<s16> new_samples(samples.size() / system_channels * device_channels, 0);
+
+                for (u32 read_index = 0, write_index = 0; read_index < samples.size();
+                     read_index += system_channels, write_index += device_channels) {
+                    const auto left_sample{static_cast<s16>(std::clamp(
+                        static_cast<s32>(
+                            static_cast<f32>(
+                                samples[read_index + static_cast<u32>(Channels::FrontLeft)]) *
+                            volume),
+                        min, max))};
+
+                    new_samples[write_index + static_cast<u32>(Channels::FrontLeft)] = left_sample;
+
+                    const auto right_sample{static_cast<s16>(std::clamp(
+                        static_cast<s32>(
+                            static_cast<f32>(
+                                samples[read_index + static_cast<u32>(Channels::FrontRight)]) *
+                            volume),
+                        min, max))};
+
+                    new_samples[write_index + static_cast<u32>(Channels::FrontRight)] =
+                        right_sample;
+                }
+                samples = std::move(new_samples);
+
+            } else if (volume != 1.0f) {
+                for (u32 i = 0; i < samples.size(); i++) {
+                    samples[i] = static_cast<s16>(std::clamp(
+                        static_cast<s32>(static_cast<f32>(samples[i]) * volume), min, max));
+                }
+            }
+
+            samples_buffer.Push(samples);
+            queue.enqueue(buffer);
+            queued_buffers++;
+        }
+    }
+
+    /**
+     * Release a buffer. Audio In only, will fill a buffer with recorded samples.
+     *
+     * @param num_samples - Maximum number of samples to receive.
+     * @return Vector of recorded samples. May have fewer than num_samples.
+     */
+    std::vector<s16> ReleaseBuffer(const u64 num_samples) override {
+        static constexpr s32 min = std::numeric_limits<s16>::min();
+        static constexpr s32 max = std::numeric_limits<s16>::max();
+
+        auto samples{samples_buffer.Pop(num_samples)};
+
+        // TODO: Up-mix to 6 channels if the game expects it.
+        // For audio input this is unlikely to ever be the case though.
+
+        // Incoming mic volume seems to always be very quiet, so multiply by an additional 8 here.
+        // TODO: Play with this and find something that works better.
+        auto volume{system_volume * device_volume * 8};
+        for (u32 i = 0; i < samples.size(); i++) {
+            samples[i] = static_cast<s16>(
+                std::clamp(static_cast<s32>(static_cast<f32>(samples[i]) * volume), min, max));
+        }
+
+        if (samples.size() < num_samples) {
+            samples.resize(num_samples, 0);
+        }
+        return samples;
+    }
+
+    /**
+     * Check if a certain buffer has been consumed (fully played).
+     *
+     * @param tag - Unique tag of a buffer to check for.
+     * @return True if the buffer has been played, otherwise false.
+     */
+    bool IsBufferConsumed(const u64 tag) override {
+        if (released_buffer.tag == 0) {
+            if (!released_buffers.try_dequeue(released_buffer)) {
+                return false;
+            }
+        }
+
+        if (released_buffer.tag == tag) {
+            released_buffer.tag = 0;
+            return true;
+        }
+        return false;
+    }
+
+    /**
+     * Empty out the buffer queue.
+     */
+    void ClearQueue() override {
+        samples_buffer.Pop();
+        while (queue.pop()) {
+        }
+        while (released_buffers.pop()) {
+        }
+        queued_buffers = 0;
+        released_buffer = {};
+        playing_buffer = {};
+        playing_buffer.consumed = true;
+    }
+
+private:
+    /**
+     * Signal events back to the audio system that a buffer was played/can be filled.
+     *
+     * @param buffer - Consumed audio buffer to be released.
+     */
+    void SignalEvent(const ::AudioCore::Sink::SinkBuffer& buffer) {
+        auto& manager{system.AudioCore().GetAudioManager()};
+        switch (type) {
+        case StreamType::Out:
+            released_buffers.enqueue(buffer);
+            manager.SetEvent(Event::Type::AudioOutManager, true);
+            break;
+        case StreamType::In:
+            released_buffers.enqueue(buffer);
+            manager.SetEvent(Event::Type::AudioInManager, true);
+            break;
+        case StreamType::Render:
+            break;
+        }
+    }
+
+    /**
+     * Main callback from Cubeb. Either expects samples from us (audio render/audio out), or will
+     * provide samples to be copied (audio in).
+     *
+     * @param stream      - Cubeb-specific data about the stream.
+     * @param user_data   - Custom data pointer passed along, points to a CubebSinkStream.
+     * @param in_buff     - Input buffer to be used if the stream is an input type.
+     * @param out_buff    - Output buffer to be used if the stream is an output type.
+     * @param num_frames_ - Number of frames of audio in the buffers. Note: Not number of samples.
+     */
+    static long DataCallback([[maybe_unused]] cubeb_stream* stream, void* user_data,
+                             [[maybe_unused]] const void* in_buff, void* out_buff,
+                             long num_frames_) {
+        auto* impl = static_cast<CubebSinkStream*>(user_data);
+        if (!impl) {
+            return -1;
+        }
+
+        const std::size_t num_channels = impl->GetDeviceChannels();
+        const std::size_t frame_size = num_channels;
+        const std::size_t frame_size_bytes = frame_size * sizeof(s16);
+        const std::size_t num_frames{static_cast<size_t>(num_frames_)};
+        size_t frames_written{0};
+        [[maybe_unused]] bool underrun{false};
+
+        if (impl->type == StreamType::In) {
+            // INPUT
+            std::span<const s16> input_buffer{reinterpret_cast<const s16*>(in_buff),
+                                              num_frames * frame_size};
+
+            while (frames_written < num_frames) {
+                auto& playing_buffer{impl->playing_buffer};
+
+                // If the playing buffer has been consumed or has no frames, we need a new one
+                if (playing_buffer.consumed || playing_buffer.frames == 0) {
+                    if (!impl->queue.try_dequeue(impl->playing_buffer)) {
+                        // If no buffer was available we've underrun, just push the samples and
+                        // continue.
+                        underrun = true;
+                        impl->samples_buffer.Push(&input_buffer[frames_written * frame_size],
+                                                  (num_frames - frames_written) * frame_size);
+                        frames_written = num_frames;
+                        continue;
+                    } else {
+                        // Successfully got a new buffer, mark the old one as consumed and signal.
+                        impl->queued_buffers--;
+                        impl->SignalEvent(impl->playing_buffer);
+                    }
+                }
+
+                // Get the minimum frames available between the currently playing buffer, and the
+                // amount we have left to fill
+                size_t frames_available{
+                    std::min(playing_buffer.frames - playing_buffer.frames_played,
+                             num_frames - frames_written)};
+
+                impl->samples_buffer.Push(&input_buffer[frames_written * frame_size],
+                                          frames_available * frame_size);
+
+                frames_written += frames_available;
+                playing_buffer.frames_played += frames_available;
+
+                // If that's all the frames in the current buffer, add its samples and mark it as
+                // consumed
+                if (playing_buffer.frames_played >= playing_buffer.frames) {
+                    impl->AddPlayedSampleCount(playing_buffer.frames_played * num_channels);
+                    impl->playing_buffer.consumed = true;
+                }
+            }
+
+            std::memcpy(&impl->last_frame[0], &input_buffer[(frames_written - 1) * frame_size],
+                        frame_size_bytes);
+        } else {
+            // OUTPUT
+            std::span<s16> output_buffer{reinterpret_cast<s16*>(out_buff), num_frames * frame_size};
+
+            while (frames_written < num_frames) {
+                auto& playing_buffer{impl->playing_buffer};
+
+                // If the playing buffer has been consumed or has no frames, we need a new one
+                if (playing_buffer.consumed || playing_buffer.frames == 0) {
+                    if (!impl->queue.try_dequeue(impl->playing_buffer)) {
+                        // If no buffer was available we've underrun, fill the remaining buffer with
+                        // the last written frame and continue.
+                        underrun = true;
+                        for (size_t i = frames_written; i < num_frames; i++) {
+                            std::memcpy(&output_buffer[i * frame_size], &impl->last_frame[0],
+                                        frame_size_bytes);
+                        }
+                        frames_written = num_frames;
+                        continue;
+                    } else {
+                        // Successfully got a new buffer, mark the old one as consumed and signal.
+                        impl->queued_buffers--;
+                        impl->SignalEvent(impl->playing_buffer);
+                    }
+                }
+
+                // Get the minimum frames available between the currently playing buffer, and the
+                // amount we have left to fill
+                size_t frames_available{
+                    std::min(playing_buffer.frames - playing_buffer.frames_played,
+                             num_frames - frames_written)};
+
+                impl->samples_buffer.Pop(&output_buffer[frames_written * frame_size],
+                                         frames_available * frame_size);
+
+                frames_written += frames_available;
+                playing_buffer.frames_played += frames_available;
+
+                // If that's all the frames in the current buffer, add its samples and mark it as
+                // consumed
+                if (playing_buffer.frames_played >= playing_buffer.frames) {
+                    impl->AddPlayedSampleCount(playing_buffer.frames_played * num_channels);
+                    impl->playing_buffer.consumed = true;
+                }
+            }
+
+            std::memcpy(&impl->last_frame[0], &output_buffer[(frames_written - 1) * frame_size],
+                        frame_size_bytes);
+        }
+
+        return num_frames_;
+    }
+
+    /**
+     * Cubeb callback for if a device state changes. Unused currently.
+     *
+     * @param stream      - Cubeb-specific data about the stream.
+     * @param user_data   - Custom data pointer passed along, points to a CubebSinkStream.
+     * @param state       - New state of the device.
+     */
+    static void StateCallback([[maybe_unused]] cubeb_stream* stream,
+                              [[maybe_unused]] void* user_data,
+                              [[maybe_unused]] cubeb_state state) {}
+
+    /// Main Cubeb context
+    cubeb* ctx{};
+    /// Cubeb stream backend
+    cubeb_stream* stream_backend{};
+    /// Name of this stream
+    std::string name{};
+    /// Type of this stream
+    StreamType type;
+    /// Core system
+    Core::System& system;
+    /// Ring buffer of the samples waiting to be played or consumed
+    Common::RingBuffer<s16, 0x10000> samples_buffer;
+    /// Audio buffers queued and waiting to play
+    Common::ReaderWriterQueue<::AudioCore::Sink::SinkBuffer> queue;
+    /// The currently-playing audio buffer
+    ::AudioCore::Sink::SinkBuffer playing_buffer{};
+    /// Audio buffers which have been played and are in queue to be released by the audio system
+    Common::ReaderWriterQueue<::AudioCore::Sink::SinkBuffer> released_buffers{};
+    /// Currently released buffer waiting to be taken by the audio system
+    ::AudioCore::Sink::SinkBuffer released_buffer{};
+    /// The last played (or received) frame of audio, used when the callback underruns
+    std::array<s16, MaxChannels> last_frame{};
+};
+
+CubebSink::CubebSink(std::string_view target_device_name) {
+    // Cubeb requires COM to be initialized on the thread calling cubeb_init on Windows
+#ifdef _WIN32
+    com_init_result = CoInitializeEx(nullptr, COINIT_MULTITHREADED);
+#endif
+
+    if (cubeb_init(&ctx, "yuzu", nullptr) != CUBEB_OK) {
+        LOG_CRITICAL(Audio_Sink, "cubeb_init failed");
+        return;
+    }
+
+    if (target_device_name != auto_device_name && !target_device_name.empty()) {
+        cubeb_device_collection collection;
+        if (cubeb_enumerate_devices(ctx, CUBEB_DEVICE_TYPE_OUTPUT, &collection) != CUBEB_OK) {
+            LOG_WARNING(Audio_Sink, "Audio output device enumeration not supported");
+        } else {
+            const auto collection_end{collection.device + collection.count};
+            const auto device{
+                std::find_if(collection.device, collection_end, [&](const cubeb_device_info& info) {
+                    return info.friendly_name != nullptr &&
+                           target_device_name == std::string(info.friendly_name);
+                })};
+            if (device != collection_end) {
+                output_device = device->devid;
+            }
+            cubeb_device_collection_destroy(ctx, &collection);
+        }
+    }
+
+    cubeb_get_max_channel_count(ctx, &device_channels);
+    device_channels = device_channels >= 6U ? 6U : 2U;
+}
+
+CubebSink::~CubebSink() {
+    if (!ctx) {
+        return;
+    }
+
+    for (auto& sink_stream : sink_streams) {
+        sink_stream.reset();
+    }
+
+    cubeb_destroy(ctx);
+
+#ifdef _WIN32
+    if (SUCCEEDED(com_init_result)) {
+        CoUninitialize();
+    }
+#endif
+}
+
+SinkStream* CubebSink::AcquireSinkStream(Core::System& system, const u32 system_channels,
+                                         const std::string& name, const StreamType type) {
+    SinkStreamPtr& stream = sink_streams.emplace_back(std::make_unique<CubebSinkStream>(
+        ctx, device_channels, system_channels, output_device, input_device, name, type, system));
+
+    return stream.get();
+}
+
+void CubebSink::CloseStream(const SinkStream* stream) {
+    for (size_t i = 0; i < sink_streams.size(); i++) {
+        if (sink_streams[i].get() == stream) {
+            sink_streams[i].reset();
+            sink_streams.erase(sink_streams.begin() + i);
+            break;
+        }
+    }
+}
+
+void CubebSink::CloseStreams() {
+    sink_streams.clear();
+}
+
+void CubebSink::PauseStreams() {
+    for (auto& stream : sink_streams) {
+        stream->Stop();
+    }
+}
+
+void CubebSink::UnpauseStreams() {
+    for (auto& stream : sink_streams) {
+        stream->Start(true);
+    }
+}
+
+f32 CubebSink::GetDeviceVolume() const {
+    if (sink_streams.empty()) {
+        return 1.0f;
+    }
+
+    return sink_streams[0]->GetDeviceVolume();
+}
+
+void CubebSink::SetDeviceVolume(const f32 volume) {
+    for (auto& stream : sink_streams) {
+        stream->SetDeviceVolume(volume);
+    }
+}
+
+void CubebSink::SetSystemVolume(const f32 volume) {
+    for (auto& stream : sink_streams) {
+        stream->SetSystemVolume(volume);
+    }
+}
+
+std::vector<std::string> ListCubebSinkDevices(const bool capture) {
+    std::vector<std::string> device_list;
+    cubeb* ctx;
+
+    if (cubeb_init(&ctx, "yuzu Device Enumerator", nullptr) != CUBEB_OK) {
+        LOG_CRITICAL(Audio_Sink, "cubeb_init failed");
+        return {};
+    }
+
+    auto type{capture ? CUBEB_DEVICE_TYPE_INPUT : CUBEB_DEVICE_TYPE_OUTPUT};
+    cubeb_device_collection collection;
+    if (cubeb_enumerate_devices(ctx, type, &collection) != CUBEB_OK) {
+        LOG_WARNING(Audio_Sink, "Audio output device enumeration not supported");
+    } else {
+        for (std::size_t i = 0; i < collection.count; i++) {
+            const cubeb_device_info& device = collection.device[i];
+            if (device.friendly_name && device.friendly_name[0] != '\0' &&
+                device.state == CUBEB_DEVICE_STATE_ENABLED) {
+                device_list.emplace_back(device.friendly_name);
+            }
+        }
+        cubeb_device_collection_destroy(ctx, &collection);
+    }
+
+    cubeb_destroy(ctx);
+    return device_list;
+}
+
+} // namespace AudioCore::Sink
diff --git a/src/audio_core/sink/cubeb_sink.h b/src/audio_core/sink/cubeb_sink.h
new file mode 100644
index 000000000..f0f43dfa1
--- /dev/null
+++ b/src/audio_core/sink/cubeb_sink.h
@@ -0,0 +1,110 @@
+// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <string>
+#include <vector>
+
+#include <cubeb/cubeb.h>
+
+#include "audio_core/sink/sink.h"
+
+namespace Core {
+class System;
+}
+
+namespace AudioCore::Sink {
+class SinkStream;
+
+/**
+ * Cubeb backend sink, holds multiple output streams and is responsible for sinking samples to
+ * hardware. Used by Audio Render, Audio In and Audio Out.
+ */
+class CubebSink final : public Sink {
+public:
+    explicit CubebSink(std::string_view device_id);
+    ~CubebSink() override;
+
+    /**
+     * Create a new sink stream.
+     *
+     * @param system          - Core system.
+     * @param system_channels - Number of channels the audio system expects.
+     *                          May differ from the device's channel count.
+     * @param name            - Name of this stream.
+     * @param type            - Type of this stream, render/in/out.
+     * @param event           - Audio render only, a signal used to prevent the renderer running too
+     *                          fast.
+     * @return A pointer to the created SinkStream
+     */
+    SinkStream* AcquireSinkStream(Core::System& system, u32 system_channels,
+                                  const std::string& name, StreamType type) override;
+
+    /**
+     * Close a given stream.
+     *
+     * @param stream - The stream to close.
+     */
+    void CloseStream(const SinkStream* stream) override;
+
+    /**
+     * Close all streams.
+     */
+    void CloseStreams() override;
+
+    /**
+     * Pause all streams.
+     */
+    void PauseStreams() override;
+
+    /**
+     * Unpause all streams.
+     */
+    void UnpauseStreams() override;
+
+    /**
+     * Get the device volume. Set from calls to the IAudioDevice service.
+     *
+     * @return Volume of the device.
+     */
+    f32 GetDeviceVolume() const override;
+
+    /**
+     * Set the device volume. Set from calls to the IAudioDevice service.
+     *
+     * @param volume - New volume of the device.
+     */
+    void SetDeviceVolume(f32 volume) override;
+
+    /**
+     * Set the system volume. Comes from the audio system using this stream.
+     *
+     * @param volume - New volume of the system.
+     */
+    void SetSystemVolume(f32 volume) override;
+
+private:
+    /// Backend Cubeb context
+    cubeb* ctx{};
+    /// Cubeb id of the actual hardware output device
+    cubeb_devid output_device{};
+    /// Cubeb id of the actual hardware input device
+    cubeb_devid input_device{};
+    /// Vector of streams managed by this sink
+    std::vector<SinkStreamPtr> sink_streams{};
+
+#ifdef _WIN32
+    /// Cubeb required COM to be initialized multi-threaded on Windows
+    u32 com_init_result = 0;
+#endif
+};
+
+/**
+ * Get a list of conencted devices from Cubeb.
+ *
+ * @param capture - Return input (capture) devices if true, otherwise output devices.
+ */
+std::vector<std::string> ListCubebSinkDevices(bool capture);
+
+} // namespace AudioCore::Sink
diff --git a/src/audio_core/sink/null_sink.h b/src/audio_core/sink/null_sink.h
new file mode 100644
index 000000000..47a342171
--- /dev/null
+++ b/src/audio_core/sink/null_sink.h
@@ -0,0 +1,52 @@
+// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include "audio_core/sink/sink.h"
+#include "audio_core/sink/sink_stream.h"
+
+namespace AudioCore::Sink {
+/**
+ * A no-op sink for when no audio out is wanted.
+ */
+class NullSink final : public Sink {
+public:
+    explicit NullSink(std::string_view) {}
+    ~NullSink() override = default;
+
+    SinkStream* AcquireSinkStream([[maybe_unused]] Core::System& system,
+                                  [[maybe_unused]] u32 system_channels,
+                                  [[maybe_unused]] const std::string& name,
+                                  [[maybe_unused]] StreamType type) override {
+        return &null_sink_stream;
+    }
+
+    void CloseStream([[maybe_unused]] const SinkStream* stream) override {}
+    void CloseStreams() override {}
+    void PauseStreams() override {}
+    void UnpauseStreams() override {}
+    f32 GetDeviceVolume() const override {
+        return 1.0f;
+    }
+    void SetDeviceVolume(f32 volume) override {}
+    void SetSystemVolume(f32 volume) override {}
+
+private:
+    struct NullSinkStreamImpl final : SinkStream {
+        void Finalize() override {}
+        void Start(bool resume = false) override {}
+        void Stop() override {}
+        void AppendBuffer([[maybe_unused]] ::AudioCore::Sink::SinkBuffer& buffer,
+                          [[maybe_unused]] std::vector<s16>& samples) override {}
+        std::vector<s16> ReleaseBuffer([[maybe_unused]] u64 num_samples) override {
+            return {};
+        }
+        bool IsBufferConsumed([[maybe_unused]] const u64 tag) {
+            return true;
+        }
+        void ClearQueue() override {}
+    } null_sink_stream;
+};
+
+} // namespace AudioCore::Sink
diff --git a/src/audio_core/sink/sdl2_sink.cpp b/src/audio_core/sink/sdl2_sink.cpp
new file mode 100644
index 000000000..d6c9ec90d
--- /dev/null
+++ b/src/audio_core/sink/sdl2_sink.cpp
@@ -0,0 +1,556 @@
+// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <algorithm>
+#include <atomic>
+
+#include "audio_core/audio_core.h"
+#include "audio_core/audio_event.h"
+#include "audio_core/audio_manager.h"
+#include "audio_core/sink/sdl2_sink.h"
+#include "audio_core/sink/sink_stream.h"
+#include "common/assert.h"
+#include "common/fixed_point.h"
+#include "common/logging/log.h"
+#include "common/reader_writer_queue.h"
+#include "common/ring_buffer.h"
+#include "common/settings.h"
+#include "core/core.h"
+
+// Ignore -Wimplicit-fallthrough due to https://github.com/libsdl-org/SDL/issues/4307
+#ifdef __clang__
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wimplicit-fallthrough"
+#endif
+#include <SDL.h>
+#ifdef __clang__
+#pragma clang diagnostic pop
+#endif
+
+namespace AudioCore::Sink {
+/**
+ * SDL sink stream, responsible for sinking samples to hardware.
+ */
+class SDLSinkStream final : public SinkStream {
+public:
+    /**
+     * Create a new sink stream.
+     *
+     * @param device_channels_ - Number of channels supported by the hardware.
+     * @param system_channels_ - Number of channels the audio systems expect.
+     * @param output_device    - Name of the output device to use for this stream.
+     * @param input_device     - Name of the input device to use for this stream.
+     * @param type_            - Type of this stream.
+     * @param system_          - Core system.
+     * @param event            - Event used only for audio renderer, signalled on buffer consume.
+     */
+    SDLSinkStream(u32 device_channels_, const u32 system_channels_,
+                  const std::string& output_device, const std::string& input_device,
+                  const StreamType type_, Core::System& system_)
+        : type{type_}, system{system_} {
+        system_channels = system_channels_;
+        device_channels = device_channels_;
+
+        SDL_AudioSpec spec;
+        spec.freq = TargetSampleRate;
+        spec.channels = static_cast<u8>(device_channels);
+        spec.format = AUDIO_S16SYS;
+        if (type == StreamType::Render) {
+            spec.samples = TargetSampleCount;
+        } else {
+            spec.samples = 1024;
+        }
+        spec.callback = &SDLSinkStream::DataCallback;
+        spec.userdata = this;
+
+        playing_buffer.consumed = true;
+
+        std::string device_name{output_device};
+        bool capture{false};
+        if (type == StreamType::In) {
+            device_name = input_device;
+            capture = true;
+        }
+
+        SDL_AudioSpec obtained;
+        if (device_name.empty()) {
+            device = SDL_OpenAudioDevice(nullptr, capture, &spec, &obtained, false);
+        } else {
+            device = SDL_OpenAudioDevice(device_name.c_str(), capture, &spec, &obtained, false);
+        }
+
+        if (device == 0) {
+            LOG_CRITICAL(Audio_Sink, "Error opening SDL audio device: {}", SDL_GetError());
+            return;
+        }
+
+        LOG_DEBUG(Service_Audio,
+                  "Opening sdl stream {} with: rate {} channels {} (system channels {}) "
+                  " samples {}",
+                  device, obtained.freq, obtained.channels, system_channels, obtained.samples);
+    }
+
+    /**
+     * Destroy the sink stream.
+     */
+    ~SDLSinkStream() override {
+        if (device == 0) {
+            return;
+        }
+
+        SDL_CloseAudioDevice(device);
+    }
+
+    /**
+     * Finalize the sink stream.
+     */
+    void Finalize() override {
+        if (device == 0) {
+            return;
+        }
+
+        SDL_CloseAudioDevice(device);
+    }
+
+    /**
+     * Start the sink stream.
+     *
+     * @param resume - Set to true if this is resuming the stream a previously-active stream.
+     *                 Default false.
+     */
+    void Start(const bool resume = false) override {
+        if (device == 0) {
+            return;
+        }
+
+        if (resume && was_playing) {
+            SDL_PauseAudioDevice(device, 0);
+            paused = false;
+        } else if (!resume) {
+            SDL_PauseAudioDevice(device, 0);
+            paused = false;
+        }
+    }
+
+    /**
+     * Stop the sink stream.
+     */
+    void Stop() {
+        if (device == 0) {
+            return;
+        }
+        SDL_PauseAudioDevice(device, 1);
+        paused = true;
+    }
+
+    /**
+     * Append a new buffer and its samples to a waiting queue to play.
+     *
+     * @param buffer  - Audio buffer information to be queued.
+     * @param samples - The s16 samples to be queue for playback.
+     */
+    void AppendBuffer(::AudioCore::Sink::SinkBuffer& buffer, std::vector<s16>& samples) override {
+        if (type == StreamType::In) {
+            queue.enqueue(buffer);
+            queued_buffers++;
+        } else {
+            constexpr s32 min = std::numeric_limits<s16>::min();
+            constexpr s32 max = std::numeric_limits<s16>::max();
+
+            auto yuzu_volume{Settings::Volume()};
+            auto volume{system_volume * device_volume * yuzu_volume};
+
+            if (system_channels == 6 && device_channels == 2) {
+                // We're given 6 channels, but our device only outputs 2, so downmix.
+                constexpr std::array<f32, 4> down_mix_coeff{1.0f, 0.707f, 0.251f, 0.707f};
+
+                for (u32 read_index = 0, write_index = 0; read_index < samples.size();
+                     read_index += system_channels, write_index += device_channels) {
+                    const auto left_sample{
+                        ((Common::FixedPoint<49, 15>(
+                              samples[read_index + static_cast<u32>(Channels::FrontLeft)]) *
+                              down_mix_coeff[0] +
+                          samples[read_index + static_cast<u32>(Channels::Center)] *
+                              down_mix_coeff[1] +
+                          samples[read_index + static_cast<u32>(Channels::LFE)] *
+                              down_mix_coeff[2] +
+                          samples[read_index + static_cast<u32>(Channels::BackLeft)] *
+                              down_mix_coeff[3]) *
+                         volume)
+                            .to_int()};
+
+                    const auto right_sample{
+                        ((Common::FixedPoint<49, 15>(
+                              samples[read_index + static_cast<u32>(Channels::FrontRight)]) *
+                              down_mix_coeff[0] +
+                          samples[read_index + static_cast<u32>(Channels::Center)] *
+                              down_mix_coeff[1] +
+                          samples[read_index + static_cast<u32>(Channels::LFE)] *
+                              down_mix_coeff[2] +
+                          samples[read_index + static_cast<u32>(Channels::BackRight)] *
+                              down_mix_coeff[3]) *
+                         volume)
+                            .to_int()};
+
+                    samples[write_index + static_cast<u32>(Channels::FrontLeft)] =
+                        static_cast<s16>(std::clamp(left_sample, min, max));
+                    samples[write_index + static_cast<u32>(Channels::FrontRight)] =
+                        static_cast<s16>(std::clamp(right_sample, min, max));
+                }
+
+                samples.resize(samples.size() / system_channels * device_channels);
+
+            } else if (system_channels == 2 && device_channels == 6) {
+                // We need moar samples! Not all games will provide 6 channel audio.
+                // TODO: Implement some upmixing here. Currently just passthrough, with other
+                // channels left as silence.
+                std::vector<s16> new_samples(samples.size() / system_channels * device_channels, 0);
+
+                for (u32 read_index = 0, write_index = 0; read_index < samples.size();
+                     read_index += system_channels, write_index += device_channels) {
+                    const auto left_sample{static_cast<s16>(std::clamp(
+                        static_cast<s32>(
+                            static_cast<f32>(
+                                samples[read_index + static_cast<u32>(Channels::FrontLeft)]) *
+                            volume),
+                        min, max))};
+
+                    new_samples[write_index + static_cast<u32>(Channels::FrontLeft)] = left_sample;
+
+                    const auto right_sample{static_cast<s16>(std::clamp(
+                        static_cast<s32>(
+                            static_cast<f32>(
+                                samples[read_index + static_cast<u32>(Channels::FrontRight)]) *
+                            volume),
+                        min, max))};
+
+                    new_samples[write_index + static_cast<u32>(Channels::FrontRight)] =
+                        right_sample;
+                }
+                samples = std::move(new_samples);
+
+            } else if (volume != 1.0f) {
+                for (u32 i = 0; i < samples.size(); i++) {
+                    samples[i] = static_cast<s16>(std::clamp(
+                        static_cast<s32>(static_cast<f32>(samples[i]) * volume), min, max));
+                }
+            }
+
+            samples_buffer.Push(samples);
+            queue.enqueue(buffer);
+            queued_buffers++;
+        }
+    }
+
+    /**
+     * Release a buffer. Audio In only, will fill a buffer with recorded samples.
+     *
+     * @param num_samples - Maximum number of samples to receive.
+     * @return Vector of recorded samples. May have fewer than num_samples.
+     */
+    std::vector<s16> ReleaseBuffer(const u64 num_samples) override {
+        static constexpr s32 min = std::numeric_limits<s16>::min();
+        static constexpr s32 max = std::numeric_limits<s16>::max();
+
+        auto samples{samples_buffer.Pop(num_samples)};
+
+        // TODO: Up-mix to 6 channels if the game expects it.
+        // For audio input this is unlikely to ever be the case though.
+
+        // Incoming mic volume seems to always be very quiet, so multiply by an additional 8 here.
+        // TODO: Play with this and find something that works better.
+        auto volume{system_volume * device_volume * 8};
+        for (u32 i = 0; i < samples.size(); i++) {
+            samples[i] = static_cast<s16>(
+                std::clamp(static_cast<s32>(static_cast<f32>(samples[i]) * volume), min, max));
+        }
+
+        if (samples.size() < num_samples) {
+            samples.resize(num_samples, 0);
+        }
+        return samples;
+    }
+
+    /**
+     * Check if a certain buffer has been consumed (fully played).
+     *
+     * @param tag - Unique tag of a buffer to check for.
+     * @return True if the buffer has been played, otherwise false.
+     */
+    bool IsBufferConsumed(const u64 tag) override {
+        if (released_buffer.tag == 0) {
+            if (!released_buffers.try_dequeue(released_buffer)) {
+                return false;
+            }
+        }
+
+        if (released_buffer.tag == tag) {
+            released_buffer.tag = 0;
+            return true;
+        }
+        return false;
+    }
+
+    /**
+     * Empty out the buffer queue.
+     */
+    void ClearQueue() override {
+        samples_buffer.Pop();
+        while (queue.pop()) {
+        }
+        while (released_buffers.pop()) {
+        }
+        released_buffer = {};
+        playing_buffer = {};
+        playing_buffer.consumed = true;
+        queued_buffers = 0;
+    }
+
+private:
+    /**
+     * Signal events back to the audio system that a buffer was played/can be filled.
+     *
+     * @param buffer - Consumed audio buffer to be released.
+     */
+    void SignalEvent(const ::AudioCore::Sink::SinkBuffer& buffer) {
+        auto& manager{system.AudioCore().GetAudioManager()};
+        switch (type) {
+        case StreamType::Out:
+            released_buffers.enqueue(buffer);
+            manager.SetEvent(Event::Type::AudioOutManager, true);
+            break;
+        case StreamType::In:
+            released_buffers.enqueue(buffer);
+            manager.SetEvent(Event::Type::AudioInManager, true);
+            break;
+        case StreamType::Render:
+            break;
+        }
+    }
+
+    /**
+     * Main callback from SDL. Either expects samples from us (audio render/audio out), or will
+     * provide samples to be copied (audio in).
+     *
+     * @param userdata - Custom data pointer passed along, points to a SDLSinkStream.
+     * @param stream   - Buffer of samples to be filled or read.
+     * @param len      - Length of the stream in bytes.
+     */
+    static void DataCallback(void* userdata, Uint8* stream, int len) {
+        auto* impl = static_cast<SDLSinkStream*>(userdata);
+
+        if (!impl) {
+            return;
+        }
+
+        const std::size_t num_channels = impl->GetDeviceChannels();
+        const std::size_t frame_size = num_channels;
+        const std::size_t frame_size_bytes = frame_size * sizeof(s16);
+        const std::size_t num_frames{len / num_channels / sizeof(s16)};
+        size_t frames_written{0};
+        [[maybe_unused]] bool underrun{false};
+
+        if (impl->type == StreamType::In) {
+            std::span<s16> input_buffer{reinterpret_cast<s16*>(stream), num_frames * frame_size};
+
+            while (frames_written < num_frames) {
+                auto& playing_buffer{impl->playing_buffer};
+
+                // If the playing buffer has been consumed or has no frames, we need a new one
+                if (playing_buffer.consumed || playing_buffer.frames == 0) {
+                    if (!impl->queue.try_dequeue(impl->playing_buffer)) {
+                        // If no buffer was available we've underrun, just push the samples and
+                        // continue.
+                        underrun = true;
+                        impl->samples_buffer.Push(&input_buffer[frames_written * frame_size],
+                                                  (num_frames - frames_written) * frame_size);
+                        frames_written = num_frames;
+                        continue;
+                    } else {
+                        impl->queued_buffers--;
+                        impl->SignalEvent(impl->playing_buffer);
+                    }
+                }
+
+                // Get the minimum frames available between the currently playing buffer, and the
+                // amount we have left to fill
+                size_t frames_available{
+                    std::min(playing_buffer.frames - playing_buffer.frames_played,
+                             num_frames - frames_written)};
+
+                impl->samples_buffer.Push(&input_buffer[frames_written * frame_size],
+                                          frames_available * frame_size);
+
+                frames_written += frames_available;
+                playing_buffer.frames_played += frames_available;
+
+                // If that's all the frames in the current buffer, add its samples and mark it as
+                // consumed
+                if (playing_buffer.frames_played >= playing_buffer.frames) {
+                    impl->AddPlayedSampleCount(playing_buffer.frames_played * num_channels);
+                    impl->playing_buffer.consumed = true;
+                }
+            }
+
+            std::memcpy(&impl->last_frame[0], &input_buffer[(frames_written - 1) * frame_size],
+                        frame_size_bytes);
+        } else {
+            std::span<s16> output_buffer{reinterpret_cast<s16*>(stream), num_frames * frame_size};
+
+            while (frames_written < num_frames) {
+                auto& playing_buffer{impl->playing_buffer};
+
+                // If the playing buffer has been consumed or has no frames, we need a new one
+                if (playing_buffer.consumed || playing_buffer.frames == 0) {
+                    if (!impl->queue.try_dequeue(impl->playing_buffer)) {
+                        // If no buffer was available we've underrun, fill the remaining buffer with
+                        // the last written frame and continue.
+                        underrun = true;
+                        for (size_t i = frames_written; i < num_frames; i++) {
+                            std::memcpy(&output_buffer[i * frame_size], &impl->last_frame[0],
+                                        frame_size_bytes);
+                        }
+                        frames_written = num_frames;
+                        continue;
+                    } else {
+                        impl->queued_buffers--;
+                        impl->SignalEvent(impl->playing_buffer);
+                    }
+                }
+
+                // Get the minimum frames available between the currently playing buffer, and the
+                // amount we have left to fill
+                size_t frames_available{
+                    std::min(playing_buffer.frames - playing_buffer.frames_played,
+                             num_frames - frames_written)};
+
+                impl->samples_buffer.Pop(&output_buffer[frames_written * frame_size],
+                                         frames_available * frame_size);
+
+                frames_written += frames_available;
+                playing_buffer.frames_played += frames_available;
+
+                // If that's all the frames in the current buffer, add its samples and mark it as
+                // consumed
+                if (playing_buffer.frames_played >= playing_buffer.frames) {
+                    impl->AddPlayedSampleCount(playing_buffer.frames_played * num_channels);
+                    impl->playing_buffer.consumed = true;
+                }
+            }
+
+            std::memcpy(&impl->last_frame[0], &output_buffer[(frames_written - 1) * frame_size],
+                        frame_size_bytes);
+        }
+    }
+
+    /// SDL device id of the opened input/output device
+    SDL_AudioDeviceID device{};
+    /// Type of this stream
+    StreamType type;
+    /// Core system
+    Core::System& system;
+    /// Ring buffer of the samples waiting to be played or consumed
+    Common::RingBuffer<s16, 0x10000> samples_buffer;
+    /// Audio buffers queued and waiting to play
+    Common::ReaderWriterQueue<::AudioCore::Sink::SinkBuffer> queue;
+    /// The currently-playing audio buffer
+    ::AudioCore::Sink::SinkBuffer playing_buffer{};
+    /// Audio buffers which have been played and are in queue to be released by the audio system
+    Common::ReaderWriterQueue<::AudioCore::Sink::SinkBuffer> released_buffers{};
+    /// Currently released buffer waiting to be taken by the audio system
+    ::AudioCore::Sink::SinkBuffer released_buffer{};
+    /// The last played (or received) frame of audio, used when the callback underruns
+    std::array<s16, MaxChannels> last_frame{};
+};
+
+SDLSink::SDLSink(std::string_view target_device_name) {
+    if (!SDL_WasInit(SDL_INIT_AUDIO)) {
+        if (SDL_InitSubSystem(SDL_INIT_AUDIO) < 0) {
+            LOG_CRITICAL(Audio_Sink, "SDL_InitSubSystem audio failed: {}", SDL_GetError());
+            return;
+        }
+    }
+
+    if (target_device_name != auto_device_name && !target_device_name.empty()) {
+        output_device = target_device_name;
+    } else {
+        output_device.clear();
+    }
+
+    device_channels = 2;
+}
+
+SDLSink::~SDLSink() = default;
+
+SinkStream* SDLSink::AcquireSinkStream(Core::System& system, const u32 system_channels,
+                                       const std::string&, const StreamType type) {
+    SinkStreamPtr& stream = sink_streams.emplace_back(std::make_unique<SDLSinkStream>(
+        device_channels, system_channels, output_device, input_device, type, system));
+    return stream.get();
+}
+
+void SDLSink::CloseStream(const SinkStream* stream) {
+    for (size_t i = 0; i < sink_streams.size(); i++) {
+        if (sink_streams[i].get() == stream) {
+            sink_streams[i].reset();
+            sink_streams.erase(sink_streams.begin() + i);
+            break;
+        }
+    }
+}
+
+void SDLSink::CloseStreams() {
+    sink_streams.clear();
+}
+
+void SDLSink::PauseStreams() {
+    for (auto& stream : sink_streams) {
+        stream->Stop();
+    }
+}
+
+void SDLSink::UnpauseStreams() {
+    for (auto& stream : sink_streams) {
+        stream->Start();
+    }
+}
+
+f32 SDLSink::GetDeviceVolume() const {
+    if (sink_streams.empty()) {
+        return 1.0f;
+    }
+
+    return sink_streams[0]->GetDeviceVolume();
+}
+
+void SDLSink::SetDeviceVolume(const f32 volume) {
+    for (auto& stream : sink_streams) {
+        stream->SetDeviceVolume(volume);
+    }
+}
+
+void SDLSink::SetSystemVolume(const f32 volume) {
+    for (auto& stream : sink_streams) {
+        stream->SetSystemVolume(volume);
+    }
+}
+
+std::vector<std::string> ListSDLSinkDevices(const bool capture) {
+    std::vector<std::string> device_list;
+
+    if (!SDL_WasInit(SDL_INIT_AUDIO)) {
+        if (SDL_InitSubSystem(SDL_INIT_AUDIO) < 0) {
+            LOG_CRITICAL(Audio_Sink, "SDL_InitSubSystem audio failed: {}", SDL_GetError());
+            return {};
+        }
+    }
+
+    const int device_count = SDL_GetNumAudioDevices(capture);
+    for (int i = 0; i < device_count; ++i) {
+        device_list.emplace_back(SDL_GetAudioDeviceName(i, 0));
+    }
+
+    return device_list;
+}
+
+} // namespace AudioCore::Sink
diff --git a/src/audio_core/sink/sdl2_sink.h b/src/audio_core/sink/sdl2_sink.h
new file mode 100644
index 000000000..186bc2fa3
--- /dev/null
+++ b/src/audio_core/sink/sdl2_sink.h
@@ -0,0 +1,101 @@
+// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <string>
+#include <vector>
+
+#include "audio_core/sink/sink.h"
+
+namespace Core {
+class System;
+}
+
+namespace AudioCore::Sink {
+class SinkStream;
+
+/**
+ * SDL backend sink, holds multiple output streams and is responsible for sinking samples to
+ * hardware. Used by Audio Render, Audio In and Audio Out.
+ */
+class SDLSink final : public Sink {
+public:
+    explicit SDLSink(std::string_view device_id);
+    ~SDLSink() override;
+
+    /**
+     * Create a new sink stream.
+     *
+     * @param system          - Core system.
+     * @param system_channels - Number of channels the audio system expects.
+     *                          May differ from the device's channel count.
+     * @param name            - Name of this stream.
+     * @param type            - Type of this stream, render/in/out.
+     * @param event           - Audio render only, a signal used to prevent the renderer running too
+     *                          fast.
+     * @return A pointer to the created SinkStream
+     */
+    SinkStream* AcquireSinkStream(Core::System& system, u32 system_channels,
+                                  const std::string& name, StreamType type) override;
+
+    /**
+     * Close a given stream.
+     *
+     * @param stream - The stream to close.
+     */
+    void CloseStream(const SinkStream* stream) override;
+
+    /**
+     * Close all streams.
+     */
+    void CloseStreams() override;
+
+    /**
+     * Pause all streams.
+     */
+    void PauseStreams() override;
+
+    /**
+     * Unpause all streams.
+     */
+    void UnpauseStreams() override;
+
+    /**
+     * Get the device volume. Set from calls to the IAudioDevice service.
+     *
+     * @return Volume of the device.
+     */
+    f32 GetDeviceVolume() const override;
+
+    /**
+     * Set the device volume. Set from calls to the IAudioDevice service.
+     *
+     * @param volume - New volume of the device.
+     */
+    void SetDeviceVolume(f32 volume) override;
+
+    /**
+     * Set the system volume. Comes from the audio system using this stream.
+     *
+     * @param volume - New volume of the system.
+     */
+    void SetSystemVolume(f32 volume) override;
+
+private:
+    /// Name of the output device used by streams
+    std::string output_device;
+    /// Name of the input device used by streams
+    std::string input_device;
+    /// Vector of streams managed by this sink
+    std::vector<SinkStreamPtr> sink_streams;
+};
+
+/**
+ * Get a list of conencted devices from Cubeb.
+ *
+ * @param capture - Return input (capture) devices if true, otherwise output devices.
+ */
+std::vector<std::string> ListSDLSinkDevices(bool capture);
+
+} // namespace AudioCore::Sink
diff --git a/src/audio_core/sink/sink.h b/src/audio_core/sink/sink.h
new file mode 100644
index 000000000..91fe455e4
--- /dev/null
+++ b/src/audio_core/sink/sink.h
@@ -0,0 +1,106 @@
+// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <memory>
+#include <string>
+
+#include "audio_core/sink/sink_stream.h"
+#include "common/common_types.h"
+
+namespace Common {
+class Event;
+}
+namespace Core {
+class System;
+}
+
+namespace AudioCore::Sink {
+
+constexpr char auto_device_name[] = "auto";
+
+/**
+ * This class is an interface for an audio sink, holds multiple output streams and is responsible
+ * for sinking samples to hardware. Used by Audio Render, Audio In and Audio Out.
+ */
+class Sink {
+public:
+    virtual ~Sink() = default;
+    /**
+     * Close a given stream.
+     *
+     * @param stream - The stream to close.
+     */
+    virtual void CloseStream(const SinkStream* stream) = 0;
+
+    /**
+     * Close all streams.
+     */
+    virtual void CloseStreams() = 0;
+
+    /**
+     * Pause all streams.
+     */
+    virtual void PauseStreams() = 0;
+
+    /**
+     * Unpause all streams.
+     */
+    virtual void UnpauseStreams() = 0;
+
+    /**
+     * Create a new sink stream, kept within this sink, with a pointer returned for use.
+     * Do not free the returned pointer. When done with the stream, call CloseStream on the sink.
+     *
+     * @param system          - Core system.
+     * @param system_channels - Number of channels the audio system expects.
+     *                          May differ from the device's channel count.
+     * @param name            - Name of this stream.
+     * @param type            - Type of this stream, render/in/out.
+     * @param event           - Audio render only, a signal used to prevent the renderer running too
+     *                          fast.
+     * @return A pointer to the created SinkStream
+     */
+    virtual SinkStream* AcquireSinkStream(Core::System& system, u32 system_channels,
+                                          const std::string& name, StreamType type) = 0;
+
+    /**
+     * Get the number of channels the hardware device supports.
+     * Either 2 or 6.
+     *
+     * @return Number of device channels.
+     */
+    u32 GetDeviceChannels() const {
+        return device_channels;
+    }
+
+    /**
+     * Get the device volume. Set from calls to the IAudioDevice service.
+     *
+     * @return Volume of the device.
+     */
+    virtual f32 GetDeviceVolume() const = 0;
+
+    /**
+     * Set the device volume. Set from calls to the IAudioDevice service.
+     *
+     * @param volume - New volume of the device.
+     */
+    virtual void SetDeviceVolume(f32 volume) = 0;
+
+    /**
+     * Set the system volume. Comes from the audio system using this stream.
+     *
+     * @param volume - New volume of the system.
+     */
+    virtual void SetSystemVolume(f32 volume) = 0;
+
+protected:
+    /// Number of device channels supported by the hardware
+    u32 device_channels{2};
+};
+
+using SinkPtr = std::unique_ptr<Sink>;
+
+} // namespace AudioCore::Sink
diff --git a/src/audio_core/sink/sink_details.cpp b/src/audio_core/sink/sink_details.cpp
new file mode 100644
index 000000000..253c0fd1e
--- /dev/null
+++ b/src/audio_core/sink/sink_details.cpp
@@ -0,0 +1,91 @@
+// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <algorithm>
+#include <memory>
+#include <string>
+#include <vector>
+#include "audio_core/sink/null_sink.h"
+#include "audio_core/sink/sink_details.h"
+#ifdef HAVE_CUBEB
+#include "audio_core/sink/cubeb_sink.h"
+#endif
+#ifdef HAVE_SDL2
+#include "audio_core/sink/sdl2_sink.h"
+#endif
+#include "common/logging/log.h"
+
+namespace AudioCore::Sink {
+namespace {
+struct SinkDetails {
+    using FactoryFn = std::unique_ptr<Sink> (*)(std::string_view);
+    using ListDevicesFn = std::vector<std::string> (*)(bool);
+
+    /// Name for this sink.
+    const char* id;
+    /// A method to call to construct an instance of this type of sink.
+    FactoryFn factory;
+    /// A method to call to list available devices.
+    ListDevicesFn list_devices;
+};
+
+// sink_details is ordered in terms of desirability, with the best choice at the top.
+constexpr SinkDetails sink_details[] = {
+#ifdef HAVE_CUBEB
+    SinkDetails{"cubeb",
+                [](std::string_view device_id) -> std::unique_ptr<Sink> {
+                    return std::make_unique<CubebSink>(device_id);
+                },
+                &ListCubebSinkDevices},
+#endif
+#ifdef HAVE_SDL2
+    SinkDetails{"sdl2",
+                [](std::string_view device_id) -> std::unique_ptr<Sink> {
+                    return std::make_unique<SDLSink>(device_id);
+                },
+                &ListSDLSinkDevices},
+#endif
+    SinkDetails{"null",
+                [](std::string_view device_id) -> std::unique_ptr<Sink> {
+                    return std::make_unique<NullSink>(device_id);
+                },
+                [](bool capture) { return std::vector<std::string>{"null"}; }},
+};
+
+const SinkDetails& GetOutputSinkDetails(std::string_view sink_id) {
+    auto iter =
+        std::find_if(std::begin(sink_details), std::end(sink_details),
+                     [sink_id](const auto& sink_detail) { return sink_detail.id == sink_id; });
+
+    if (sink_id == "auto" || iter == std::end(sink_details)) {
+        if (sink_id != "auto") {
+            LOG_ERROR(Audio, "AudioCore::Sink::GetOutputSinkDetails given invalid sink_id {}",
+                      sink_id);
+        }
+        // Auto-select.
+        // sink_details is ordered in terms of desirability, with the best choice at the front.
+        iter = std::begin(sink_details);
+    }
+
+    return *iter;
+}
+} // Anonymous namespace
+
+std::vector<const char*> GetSinkIDs() {
+    std::vector<const char*> sink_ids(std::size(sink_details));
+
+    std::transform(std::begin(sink_details), std::end(sink_details), std::begin(sink_ids),
+                   [](const auto& sink) { return sink.id; });
+
+    return sink_ids;
+}
+
+std::vector<std::string> GetDeviceListForSink(std::string_view sink_id, bool capture) {
+    return GetOutputSinkDetails(sink_id).list_devices(capture);
+}
+
+std::unique_ptr<Sink> CreateSinkFromID(std::string_view sink_id, std::string_view device_id) {
+    return GetOutputSinkDetails(sink_id).factory(device_id);
+}
+
+} // namespace AudioCore::Sink
diff --git a/src/audio_core/sink/sink_details.h b/src/audio_core/sink/sink_details.h
new file mode 100644
index 000000000..3ebdb1e30
--- /dev/null
+++ b/src/audio_core/sink/sink_details.h
@@ -0,0 +1,43 @@
+// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <string>
+#include <string_view>
+#include <vector>
+
+namespace AudioCore {
+class AudioManager;
+
+namespace Sink {
+
+class Sink;
+
+/**
+ * Retrieves the IDs for all available audio sinks.
+ *
+ * @return Vector of available sink names.
+ */
+std::vector<const char*> GetSinkIDs();
+
+/**
+ * Gets the list of devices for a particular sink identified by the given ID.
+ *
+ * @param sink_id - Id of the sink to get devices from.
+ * @param capture - Get capture (input) devices, or output devices?
+ * @return Vector of device names.
+ */
+std::vector<std::string> GetDeviceListForSink(std::string_view sink_id, bool capture);
+
+/**
+ * Creates an audio sink identified by the given device ID.
+ *
+ * @param sink_id   - Id of the sink to create.
+ * @param device_id - Name of the device to create.
+ * @return Pointer to the created sink.
+ */
+std::unique_ptr<Sink> CreateSinkFromID(std::string_view sink_id, std::string_view device_id);
+
+} // namespace Sink
+} // namespace AudioCore
diff --git a/src/audio_core/sink/sink_stream.h b/src/audio_core/sink/sink_stream.h
new file mode 100644
index 000000000..17ed6593f
--- /dev/null
+++ b/src/audio_core/sink/sink_stream.h
@@ -0,0 +1,224 @@
+// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <atomic>
+#include <memory>
+#include <vector>
+
+#include "audio_core/common/common.h"
+#include "common/common_types.h"
+
+namespace AudioCore::Sink {
+
+enum class StreamType {
+    Render,
+    Out,
+    In,
+};
+
+struct SinkBuffer {
+    u64 frames;
+    u64 frames_played;
+    u64 tag;
+    bool consumed;
+};
+
+/**
+ * Contains a real backend stream for outputting samples to hardware,
+ * created only via a Sink (See Sink::AcquireSinkStream).
+ *
+ * Accepts a SinkBuffer and samples in PCM16 format to be output (see AppendBuffer).
+ * Appended buffers act as a FIFO queue, and will be held until played.
+ * You should regularly call IsBufferConsumed with the unique SinkBuffer tag to check if the buffer
+ * has been consumed.
+ *
+ * Since these are a FIFO queue, always check IsBufferConsumed in the same order you appended the
+ * buffers, skipping a buffer will result in all following buffers to never release.
+ *
+ * If the buffers appear to be stuck, you can stop and re-open an IAudioIn/IAudioOut service (this
+ * is what games do), or call ClearQueue to flush all of the buffers without a full restart.
+ */
+class SinkStream {
+public:
+    virtual ~SinkStream() = default;
+
+    /**
+     * Finalize the sink stream.
+     */
+    virtual void Finalize() = 0;
+
+    /**
+     * Start the sink stream.
+     *
+     * @param resume - Set to true if this is resuming the stream a previously-active stream.
+     *                 Default false.
+     */
+    virtual void Start(bool resume = false) = 0;
+
+    /**
+     * Stop the sink stream.
+     */
+    virtual void Stop() = 0;
+
+    /**
+     * Append a new buffer and its samples to a waiting queue to play.
+     *
+     * @param buffer  - Audio buffer information to be queued.
+     * @param samples - The s16 samples to be queue for playback.
+     */
+    virtual void AppendBuffer(SinkBuffer& buffer, std::vector<s16>& samples) = 0;
+
+    /**
+     * Release a buffer. Audio In only, will fill a buffer with recorded samples.
+     *
+     * @param num_samples - Maximum number of samples to receive.
+     * @return Vector of recorded samples. May have fewer than num_samples.
+     */
+    virtual std::vector<s16> ReleaseBuffer(u64 num_samples) = 0;
+
+    /**
+     * Check if a certain buffer has been consumed (fully played).
+     *
+     * @param tag - Unique tag of a buffer to check for.
+     * @return True if the buffer has been played, otherwise false.
+     */
+    virtual bool IsBufferConsumed(u64 tag) = 0;
+
+    /**
+     * Empty out the buffer queue.
+     */
+    virtual void ClearQueue() = 0;
+
+    /**
+     * Check if the stream is paused.
+     *
+     * @return True if paused, otherwise false.
+     */
+    bool IsPaused() {
+        return paused;
+    }
+
+    /**
+     * Get the number of system channels in this stream.
+     *
+     * @return Number of system channels.
+     */
+    u32 GetSystemChannels() const {
+        return system_channels;
+    }
+
+    /**
+     * Set the number of channels the system expects.
+     *
+     * @param channels - New number of system channels.
+     */
+    void SetSystemChannels(u32 channels) {
+        system_channels = channels;
+    }
+
+    /**
+     * Get the number of channels the hardware supports.
+     *
+     * @return Number of channels supported.
+     */
+    u32 GetDeviceChannels() const {
+        return device_channels;
+    }
+
+    /**
+     * Get the total number of samples played by this stream.
+     *
+     * @return Number of samples played.
+     */
+    u64 GetPlayedSampleCount() const {
+        return played_sample_count;
+    }
+
+    /**
+     * Set the number of samples played.
+     * This is started and stopped on system start/stop.
+     *
+     * @param played_sample_count_ - Number of samples to set.
+     */
+    void SetPlayedSampleCount(u64 played_sample_count_) {
+        played_sample_count = played_sample_count_;
+    }
+
+    /**
+     * Add to the played sample count.
+     *
+     * @param num_samples - Number of samples to add.
+     */
+    void AddPlayedSampleCount(u64 num_samples) {
+        played_sample_count += num_samples;
+    }
+
+    /**
+     * Get the system volume.
+     *
+     * @return The current system volume.
+     */
+    f32 GetSystemVolume() const {
+        return system_volume;
+    }
+
+    /**
+     * Get the device volume.
+     *
+     * @return The current device volume.
+     */
+    f32 GetDeviceVolume() const {
+        return device_volume;
+    }
+
+    /**
+     * Set the system volume.
+     *
+     * @param volume_ - The new system volume.
+     */
+    void SetSystemVolume(f32 volume_) {
+        system_volume = volume_;
+    }
+
+    /**
+     * Set the device volume.
+     *
+     * @param volume_ - The new device volume.
+     */
+    void SetDeviceVolume(f32 volume_) {
+        device_volume = volume_;
+    }
+
+    /**
+     * Get the number of queued audio buffers.
+     *
+     * @return The number of queued buffers.
+     */
+    u32 GetQueueSize() {
+        return queued_buffers.load();
+    }
+
+protected:
+    /// Number of buffers waiting to be played
+    std::atomic<u32> queued_buffers{};
+    /// Total samples played by this stream
+    std::atomic<u64> played_sample_count{};
+    /// Set by the audio render/in/out system which uses this stream
+    f32 system_volume{1.0f};
+    /// Set via IAudioDevice service calls
+    f32 device_volume{1.0f};
+    /// Set by the audio render/in/out systen which uses this stream
+    u32 system_channels{2};
+    /// Channels supported by hardware
+    u32 device_channels{2};
+    /// Is this stream currently paused?
+    std::atomic<bool> paused{true};
+    /// Was this stream previously playing?
+    std::atomic<bool> was_playing{false};
+};
+
+using SinkStreamPtr = std::unique_ptr<SinkStream>;
+
+} // namespace AudioCore::Sink