diff --git a/src/audio/CMakeLists.txt b/src/audio/CMakeLists.txt
index 4986a7da..57dae1cd 100644
--- a/src/audio/CMakeLists.txt
+++ b/src/audio/CMakeLists.txt
@@ -2,6 +2,7 @@ idf_component_register(
   SRCS "audio_decoder.cpp" "audio_task.cpp" "chunk.cpp" "fatfs_audio_input.cpp"
   "stream_info.cpp" "stream_message.cpp" "i2s_audio_output.cpp"
   "stream_buffer.cpp" "audio_playback.cpp" "audio_element_handle.cpp"
+  "stream_event.cpp" "audio_element.cpp"
   INCLUDE_DIRS "include"
   REQUIRES "codecs" "drivers" "cbor" "result" "tasks" "span")
 
diff --git a/src/audio/audio_decoder.cpp b/src/audio/audio_decoder.cpp
index 0b3d9878..8ef90905 100644
--- a/src/audio/audio_decoder.cpp
+++ b/src/audio/audio_decoder.cpp
@@ -4,6 +4,7 @@
 
 #include <cstddef>
 #include <cstdint>
+#include <memory>
 
 #include "freertos/FreeRTOS.h"
 
@@ -14,19 +15,30 @@
 #include "audio_element.hpp"
 #include "chunk.hpp"
 #include "fatfs_audio_input.hpp"
-
-static const char* kTag = "DEC";
+#include "stream_info.hpp"
 
 namespace audio {
 
+static const std::size_t kSamplesPerChunk = 256;
+
 AudioDecoder::AudioDecoder() : IAudioElement(), stream_info_({}) {}
 
 AudioDecoder::~AudioDecoder() {}
 
+auto AudioDecoder::HasUnprocessedInput() -> bool {
+  return !needs_more_input_ || has_samples_to_send_;
+}
+
 auto AudioDecoder::ProcessStreamInfo(const StreamInfo& info)
     -> cpp::result<void, AudioProcessingError> {
   stream_info_ = info;
 
+  if (info.ChunkSize()) {
+    chunk_reader_.emplace(info.ChunkSize().value());
+  } else {
+    // TODO.
+  }
+
   // Reuse the existing codec if we can. This will help with gapless playback,
   // since we can potentially just continue to decode as we were before,
   // without any setup overhead.
@@ -42,71 +54,66 @@ auto AudioDecoder::ProcessStreamInfo(const StreamInfo& info)
     return cpp::fail(UNSUPPORTED_STREAM);
   }
 
+  // TODO: defer until first header read, so we can give better info about
+  // sample rate, chunk size, etc.
+  auto downstream_info = StreamEvent::CreateStreamInfo(
+      input_events_, std::make_unique<StreamInfo>(info));
+  downstream_info->stream_info->BitsPerSample(32);
+  downstream_info->stream_info->SampleRate(48'000);
+  chunk_size_ = 128;
+  downstream_info->stream_info->ChunkSize(chunk_size_);
+
+  SendOrBufferEvent(std::move(downstream_info));
+
   return {};
 }
 
 auto AudioDecoder::ProcessChunk(const cpp::span<std::byte>& chunk)
     -> cpp::result<size_t, AudioProcessingError> {
-  if (current_codec_ == nullptr) {
+  if (current_codec_ == nullptr || !chunk_reader_) {
     // Should never happen, but fail explicitly anyway.
     return cpp::fail(UNSUPPORTED_STREAM);
   }
 
-  current_codec_->SetInput(chunk);
-
-  bool has_samples_to_send = false;
-  bool needs_more_input = false;
-  std::optional<codecs::ICodec::ProcessingError> error = std::nullopt;
-  while (1) {
-    ChunkWriteResult res = chunk_writer_->WriteChunkToStream(
-        [&](cpp::span<std::byte> buffer) -> std::size_t {
-          std::size_t bytes_written = 0;
-          // Continue filling up the output buffer so long as we have samples
-          // leftover, or are able to synthesize more samples from the input.
-          while (has_samples_to_send || !needs_more_input) {
-            if (!has_samples_to_send) {
-              auto result = current_codec_->ProcessNextFrame();
-              has_samples_to_send = true;
-              if (result.has_error()) {
-                error = result.error();
-                // End our output stream immediately if the codec barfed.
-                return 0;
-              } else {
-                needs_more_input = result.value();
-              }
-            } else {
-              auto result = current_codec_->WriteOutputSamples(
-                  buffer.last(buffer.size() - bytes_written));
-              bytes_written += result.first;
-              has_samples_to_send = !result.second;
-            }
-          }
-          return bytes_written;
-        },
-        // TODO
-        portMAX_DELAY);
-
-    switch (res) {
-      case CHUNK_WRITE_OKAY:
-        break;
-      case CHUNK_WRITE_TIMEOUT:
-      case CHUNK_OUT_OF_DATA:
+  current_codec_->SetInput(chunk_reader_->HandleNewData(chunk));
+
+  return {};
+}
+
+auto AudioDecoder::Process() -> cpp::result<void, AudioProcessingError> {
+  if (has_samples_to_send_) {
+    // Writing samples is relatively quick (it's just a bunch of memcopy's), so
+    // do them all at once.
+    while (has_samples_to_send_ && !IsOverBuffered()) {
+      auto buffer = StreamEvent::CreateChunkData(input_events_, chunk_size_);
+      auto write_res =
+          current_codec_->WriteOutputSamples(buffer->chunk_data.bytes);
+      buffer->chunk_data.bytes =
+          buffer->chunk_data.bytes.first(write_res.first);
+      has_samples_to_send_ = !write_res.second;
+
+      if (!SendOrBufferEvent(std::move(buffer))) {
         return {};
-      default:
-        return cpp::fail(IO_ERROR);
+      }
     }
+    // We will process the next frame during the next call to this method.
+    return {};
   }
 
-  if (error) {
-    ESP_LOGE(kTag, "Codec encountered error %d", error.value());
-    return cpp::fail(IO_ERROR);
-  }
+  if (!needs_more_input_) {
+    auto res = current_codec_->ProcessNextFrame();
+    if (res.has_error()) {
+      // todo
+      return {};
+    }
+    needs_more_input_ = res.value();
+    has_samples_to_send_ = true;
 
-  return current_codec_->GetInputPosition();
-}
+    if (needs_more_input_) {
+      chunk_reader_->HandleLeftovers(current_codec_->GetInputPosition());
+    }
+  }
 
-auto AudioDecoder::ProcessIdle() -> cpp::result<void, AudioProcessingError> {
-  // Not used; we delay forever when waiting on IO.
   return {};
 }
 
diff --git a/src/audio/audio_element.cpp b/src/audio/audio_element.cpp
new file mode 100644
index 00000000..e1623c36
--- /dev/null
+++ b/src/audio/audio_element.cpp
@@ -0,0 +1,59 @@
+#include "audio_element.hpp"
+
+namespace audio {
+
+IAudioElement::IAudioElement()
+    : input_events_(xQueueCreate(kEventQueueSize, sizeof(StreamEvent))),
+      output_events_(nullptr),
+      unprocessed_output_chunks_(0),
+      buffered_output_(),
+      current_state_(STATE_RUN) {}
+
+IAudioElement::~IAudioElement() {
+  // Ensure we don't leak any memory from events leftover in the queue.
+  while (uxQueueSpacesAvailable(input_events_) < kEventQueueSize) {
+    StreamEvent* event;
+    if (xQueueReceive(input_events_, &event, 0)) {
+      free(event);
+    } else {
+      break;
+    }
+  }
+  // Technically there's a race here if someone is still adding to the queue,
+  // but hopefully the whole pipeline is stopped if an element is being
+  // destroyed.
+  vQueueDelete(input_events_);
+}
+
+auto IAudioElement::SendOrBufferEvent(std::unique_ptr<StreamEvent> event)
+    -> bool {
+  if (event->tag == StreamEvent::CHUNK_DATA) {
+    unprocessed_output_chunks_++;
+  }
+  if (!buffered_output_.empty()) {
+    // To ensure we send data in order, don't try to send if we've already
+    // failed to send something.
+    buffered_output_.push_back(std::move(event));
+    return false;
+  }
+  StreamEvent* raw_event = event.release();
+  if (!xQueueSend(output_events_, raw_event, 0)) {
+    buffered_output_.emplace_front(raw_event);
+    return false;
+  }
+  return true;
+}
+
+auto IAudioElement::FlushBufferedOutput() -> bool {
+  while (!buffered_output_.empty()) {
+    StreamEvent* raw_event = buffered_output_.front().release();
+    buffered_output_.pop_front();
+    if (!xQueueSend(output_events_, raw_event, 0)) {
+      buffered_output_.emplace_front(raw_event);
+      return false;
+    }
+  }
+  return true;
+}
+
+}  // namespace audio
diff --git a/src/audio/audio_playback.cpp b/src/audio/audio_playback.cpp
index 7462b4f6..fae096d0 100644
--- a/src/audio/audio_playback.cpp
+++ b/src/audio/audio_playback.cpp
@@ -18,9 +18,6 @@
 
 namespace audio {
 
-// TODO: idk
-static const std::size_t kMinElementBufferSize = 1024;
-
 auto AudioPlayback::create(drivers::GpioExpander* expander,
                            std::shared_ptr<drivers::SdStorage> storage)
     -> cpp::result<std::unique_ptr<AudioPlayback>, Error> {
@@ -37,8 +34,6 @@ auto AudioPlayback::create(drivers::GpioExpander* expander,
   auto playback = std::make_unique<AudioPlayback>();
 
   // Configure the pipeline
-  source->InputBuffer(&playback->stream_start_);
-  sink->OutputBuffer(&playback->stream_end_);
   playback->ConnectElements(source.get(), codec.get());
   playback->ConnectElements(codec.get(), sink.get());
 
@@ -52,9 +47,7 @@ auto AudioPlayback::create(drivers::GpioExpander* expander,
   return playback;
 }
 
-// TODO(jacqueline): think about sizes
-AudioPlayback::AudioPlayback()
-    : stream_start_(128, 256), stream_end_(128, 256) {}
+AudioPlayback::AudioPlayback() {}
 
 AudioPlayback::~AudioPlayback() {
   for (auto& element : element_handles_) {
@@ -63,33 +56,16 @@ AudioPlayback::~AudioPlayback() {
 }
 
 auto AudioPlayback::Play(const std::string& filename) -> void {
-  StreamInfo info;
-  info.Path(filename);
-
-  std::array<std::byte, 128> dest;
-  auto len = WriteMessage(
-      TYPE_STREAM_INFO, [&](auto enc) { return info.Encode(enc); }, dest);
+  auto info = std::make_unique<StreamInfo>();
+  info->Path(filename);
+  auto event = StreamEvent::CreateStreamInfo(nullptr, std::move(info));
 
-  if (len.has_error()) {
-    // TODO.
-    return;
-  }
-
-  // TODO: short delay, return error on fail
-  xMessageBufferSend(*stream_start_.Handle(), dest.data(), len.value(),
-                     portMAX_DELAY);
+  xQueueSend(input_handle_, event.release(), portMAX_DELAY);
 }
 
 auto AudioPlayback::ConnectElements(IAudioElement* src, IAudioElement* sink)
     -> void {
-  std::size_t chunk_size =
-      std::max(src->InputMinChunkSize(), sink->InputMinChunkSize());
-  std::size_t buffer_size = std::max(kMinElementBufferSize, chunk_size * 2);
-
-  auto buffer = std::make_unique<StreamBuffer>(chunk_size, buffer_size);
-  src->OutputBuffer(buffer.get());
-  sink->OutputBuffer(buffer.get());
-  element_buffers_.push_back(std::move(buffer));
+  src->OutputEventQueue(sink->InputEventQueue());
 }
 
 }  // namespace audio
diff --git a/src/audio/audio_task.cpp b/src/audio/audio_task.cpp
index e3641973..f0a01b6c 100644
--- a/src/audio/audio_task.cpp
+++ b/src/audio/audio_task.cpp
@@ -3,6 +3,7 @@
 #include <stdlib.h>
 
 #include <cstdint>
+#include <deque>
 #include <memory>
 
 #include "audio_element_handle.hpp"
@@ -10,19 +11,21 @@
 #include "esp_heap_caps.h"
 #include "esp_log.h"
 #include "freertos/portmacro.h"
+#include "freertos/projdefs.h"
 #include "freertos/queue.h"
-#include "freertos/stream_buffer.h"
 #include "span.hpp"
 
 #include "audio_element.hpp"
 #include "chunk.hpp"
+#include "stream_event.hpp"
 #include "stream_info.hpp"
 #include "stream_message.hpp"
+#include "sys/_stdint.h"
 #include "tasks.hpp"
 
 namespace audio {
 
-  static const char *kTag = "task";
+static const char* kTag = "task";
 
 auto StartAudioTask(const std::string& name,
                     std::shared_ptr<IAudioElement> element)
@@ -47,84 +50,91 @@ void AudioTaskMain(void* args) {
     std::shared_ptr<IAudioElement> element = std::move(real_args->element);
     delete real_args;
 
-    ChunkReader chunk_reader = ChunkReader(element->InputBuffer());
+    // Queue of events that we have received on our input queue, but not yet
+    // processed.
+    std::deque<std::unique_ptr<StreamEvent>> pending_events;
 
     while (element->ElementState() != STATE_QUIT) {
-      if (element->ElementState() == STATE_PAUSE) {
-        // TODO: park with a condition variable or something?
-        vTaskDelay(1000);
+      // First, we pull events from our input queue into pending_events. This
+      // keeps us responsive to any events that need to be handled immediately.
+      // Then we check if there's any events to flush downstream.
+      // Then we pass anything requiring processing to the element.
+
+      bool has_work_to_do =
+          (!pending_events.empty() || element->HasUnflushedOutput() ||
+           element->HasUnprocessedInput()) &&
+          !element->IsOverBuffered();
+
+      // If we have no new events to process and the element has nothing left to
+      // do, then just delay forever waiting for a new event.
+      TickType_t ticks_to_wait = has_work_to_do ? 0 : portMAX_DELAY;
+
+      StreamEvent* event_ptr = nullptr;
+      bool has_event =
+          xQueueReceive(element->InputEventQueue(), &event_ptr, ticks_to_wait);
+
+      if (has_event && event_ptr != nullptr) {
+        std::unique_ptr<StreamEvent> event(event_ptr);
+        if (event->tag == StreamEvent::CHUNK_NOTIFICATION) {
+          element->OnChunkProcessed();
+        } else {
+          // This isn't an event that needs to be actioned immediately. Add it
+          // to our work queue.
+          pending_events.push_back(std::move(event));
+        }
+        // Loop again, so that we service all incoming events before doing our
+        // possibly expensive processing.
         continue;
       }
 
-      cpp::result<size_t, AudioProcessingError> process_res;
-
-      // If this element has an input stream, then our top priority is
-      // processing any chunks from it. Try doing this first, then fall back to
-      // the other cases.
-      bool has_received_message = false;
-      ChunkReadResult chunk_res = chunk_reader.ReadChunkFromStream(
-          [&](cpp::span<std::byte> data) -> std::optional<size_t> {
-            process_res = element->ProcessChunk(data);
-            if (process_res.has_value()) {
-              return process_res.value();
-            } else {
-              return {};
-            }
-          },
-          0);
-
-      if (chunk_res == CHUNK_PROCESSING_ERROR ||
-          chunk_res == CHUNK_DECODING_ERROR) {
-        ESP_LOGE(kTag, "failed to process chunk");
-        break;  // TODO.
-      } else if (chunk_res == CHUNK_STREAM_ENDED) {
-        has_received_message = true;
+      // We have no new events. Next, see if there's anything that needs to be
+      // flushed.
+      if (element->HasUnflushedOutput() && !element->FlushBufferedOutput()) {
+        // We had things to flush, but couldn't send it all. This probably
+        // implies that the downstream element is having issues servicing its
+        // input queue, so hold off for a moment before retrying.
+        ESP_LOGW(kTag, "failed to flush buffered output");
+        vTaskDelay(pdMS_TO_TICKS(100));
+        continue;
       }
 
-      if (has_received_message) {
-        auto message = chunk_reader.GetLastMessage();
-        MessageType type = ReadMessageType(message);
-        if (type == TYPE_STREAM_INFO) {
-          auto parse_res = ReadMessage<StreamInfo>(&StreamInfo::Parse, message);
-          if (parse_res.has_error()) {
-            ESP_LOGE(kTag, "failed to parse stream info");
-            break;  // TODO.
-          }
-          auto info_res = element->ProcessStreamInfo(parse_res.value());
-          if (info_res.has_error()) {
-            ESP_LOGE(kTag, "failed to process stream info");
-            break;  // TODO.
-          }
+      if (element->HasUnprocessedInput()) {
+        auto process_res = element->Process();
+        if (!process_res.has_error() || process_res.error() != OUT_OF_DATA) {
+          // TODO: log!
         }
+        continue;
       }
 
-      // Chunk reading must have timed out, or we don't have an input stream.
-      ElementState state = element->ElementState();
-      if (state == STATE_PAUSE) {
-        element->PrepareForPause();
-
-        vTaskSuspend(NULL);
+      // The element ran out of data, so now it's time to let it process more
+      // input.
+      while (!pending_events.empty()) {
+        auto event = std::move(pending_events.front());
+        pending_events.pop_front();
 
-        // Zzzzzz...
-
-        // When we wake up, skip straight to the start of the loop again.
-        continue;
-      } else if (state == STATE_QUIT) {
-        break;
-      }
+        if (event->tag == StreamEvent::STREAM_INFO) {
+          auto process_res = element->ProcessStreamInfo(*event->stream_info);
+          if (process_res.has_error()) {
+            // TODO(jacqueline)
+            ESP_LOGE(kTag, "failed to process stream info");
+          }
+        } else if (event->tag == StreamEvent::CHUNK_DATA) {
+          StreamEvent* callback = new StreamEvent();
+          callback->source = element->InputEventQueue();
+          callback->tag = StreamEvent::CHUNK_NOTIFICATION;
+          if (!xQueueSend(event->source, callback, 0)) {
+            // TODO: log? crash? hmm.
+            pending_events.push_front(std::move(event));
+            continue;
+          }
 
-      // Signal the element to do any of its idle tasks.
-      auto process_error = element->ProcessIdle();
-      if (process_error.has_error()) {
-        auto err = process_error.error();
-        if (err == OUT_OF_DATA) {
-          // If we ran out of data, then place ourselves into the pause state.
-          // We will be woken up when there's something to do.
-          element->ElementState(STATE_PAUSE);
-          continue;
-        } else {
-          ESP_LOGE(kTag, "failed to process idle");
-          break;  // TODO.
+          auto process_chunk_res =
+              element->ProcessChunk(event->chunk_data.bytes);
+          if (process_chunk_res.has_error()) {
+            // TODO(jacqueline)
+            ESP_LOGE(kTag, "failed to process chunk");
+            continue;
+          }
         }
       }
     }
diff --git a/src/audio/chunk.cpp b/src/audio/chunk.cpp
index b37fdfed..baf2aba5 100644
--- a/src/audio/chunk.cpp
+++ b/src/audio/chunk.cpp
@@ -14,130 +14,37 @@
 
 namespace audio {
 
-static const char* kTag = "chunk";
+ChunkReader::ChunkReader(std::size_t chunk_size)
+    : raw_working_buffer_(static_cast<std::byte*>(
+          heap_caps_malloc(chunk_size * 1.5, MALLOC_CAP_SPIRAM))),
+      working_buffer_(raw_working_buffer_, chunk_size * 1.5) {}
 
-ChunkWriter::ChunkWriter(StreamBuffer* buffer)
-    : stream_(buffer), leftover_bytes_(0) {}
-
-ChunkWriter::~ChunkWriter() {}
-
-auto ChunkWriter::Reset() -> void {
-  leftover_bytes_ = 0;
-}
-
-auto ChunkWriter::WriteChunkToStream(
-    std::function<size_t(cpp::span<std::byte>)> callback,
-    TickType_t max_wait) -> ChunkWriteResult {
-  cpp::span<std::byte> write_buffer = stream_->WriteBuffer();
-  // First, write out our chunk header so we know how much space to give to
-  // the callback.
-  auto header_size = WriteTypeOnlyMessage(TYPE_CHUNK_HEADER, write_buffer);
-  if (header_size.has_error()) {
-    return CHUNK_ENCODING_ERROR;
-  }
-
-  // Now we can ask the callback to fill the remaining space. If the previous
-  // call to this method timed out, then we may already have the data we need
-  // in our write buffer.
-  size_t chunk_size;
-  if (leftover_bytes_ > 0) {
-    chunk_size = leftover_bytes_;
-  } else {
-    chunk_size = std::invoke(
-        callback,
-        write_buffer.subspan(header_size.value(),
-                             write_buffer.size() - header_size.value()));
-  }
-
-  if (chunk_size == 0) {
-    // They had nothing for us, so bail out.
-    return CHUNK_OUT_OF_DATA;
-  }
-
-  // Try to write to the buffer. Note the return type here will be either 0 or
-  // header_size + chunk_size, as MessageBuffer doesn't allow partial writes.
-  size_t intended_write_size = header_size.value() + chunk_size;
-  ESP_LOGI(kTag, "writing chunk of size %d", intended_write_size);
-  size_t actual_write_size =
-      xMessageBufferSend(stream_->Handle(), write_buffer.data(),
-          intended_write_size, max_wait);
-
-  if (actual_write_size == 0) {
-    leftover_bytes_ = chunk_size;
-    return CHUNK_WRITE_TIMEOUT;
-  } else {
-    leftover_bytes_ = 0;
-  }
-
-  return CHUNK_WRITE_OKAY;
+ChunkReader::~ChunkReader() {
+  free(raw_working_buffer_);
 }
 
-ChunkReader::ChunkReader(StreamBuffer* stream) : stream_(stream) {}
-
-ChunkReader::~ChunkReader() {}
-
-auto ChunkReader::Reset() -> void {
+auto ChunkReader::HandleNewData(cpp::span<std::byte> data)
+    -> cpp::span<std::byte> {
+  // Copy the new data onto the front for anything that was left over from the
+  // last portion. Note: this could be optimised for the '0 leftover bytes'
+  // case, which technically shouldn't need a copy.
+  cpp::span<std::byte> new_data_dest = working_buffer_.subspan(leftover_bytes_);
+  std::copy(data.begin(), data.end(), new_data_dest.begin());
+  last_data_in_working_buffer_ =
+      working_buffer_.first(leftover_bytes_ + data.size());
   leftover_bytes_ = 0;
-  last_message_size_ = 0;
-}
-
-auto ChunkReader::GetLastMessage() -> cpp::span<std::byte> {
-  return stream_->ReadBuffer().subspan(leftover_bytes_, last_message_size_);
+  return last_data_in_working_buffer_;
 }
 
-auto ChunkReader::ReadChunkFromStream(
-    std::function<std::optional<size_t>(cpp::span<std::byte>)> callback,
-    TickType_t max_wait) -> ChunkReadResult {
-  // First, wait for a message to arrive over the buffer.
-  cpp::span<std::byte> new_data_dest = stream_->ReadBuffer().last(
-      stream_->ReadBuffer().size() - leftover_bytes_);
-  ESP_LOGI(kTag, "reading chunk of size %d", new_data_dest.size());
-  last_message_size_ = xMessageBufferReceive(
-      stream_->Handle(), new_data_dest.data(), new_data_dest.size(), max_wait);
-
-  if (last_message_size_ == 0) {
-    return CHUNK_READ_TIMEOUT;
-  }
-
-  cpp::span<std::byte> new_data = GetLastMessage();
-  MessageType type = ReadMessageType(new_data);
-
-  if (type != TYPE_CHUNK_HEADER) {
-    // This message wasn't for us, so let the caller handle it.
-    Reset();
-    return CHUNK_STREAM_ENDED;
-  }
-
-  // Work the size and position of the chunk.
-  auto chunk_data = GetAdditionalData(new_data);
-
-  // Now we need to stick the end of the last chunk (if it exists) onto the
-  // front of the new chunk. Do it this way around bc we assume the old chunk
-  // is shorter, and therefore faster to move.
-  cpp::span<std::byte> leftover_data =
-      stream_->ReadBuffer().first(leftover_bytes_);
-  cpp::span<std::byte> combined_data(chunk_data.data() - leftover_data.size(),
-                                     leftover_data.size() + chunk_data.size());
-  if (leftover_bytes_ > 0) {
-    std::copy_backward(leftover_data.begin(), leftover_data.end(),
-                       combined_data.begin());
-  }
-
-  // Tell the callback about the new data.
-  std::optional<size_t> amount_processed = std::invoke(callback, combined_data);
-  if (!amount_processed) {
-    return CHUNK_PROCESSING_ERROR;
-  }
-
-  // Prepare for the next iteration.
-  leftover_bytes_ = combined_data.size() - amount_processed.value();
+auto ChunkReader::HandleLeftovers(std::size_t bytes_used) -> void {
+  leftover_bytes_ = last_data_in_working_buffer_.size() - bytes_used;
   if (leftover_bytes_ > 0) {
-    std::copy(combined_data.begin() + amount_processed.value(),
-              combined_data.end(), stream_->ReadBuffer().begin());
-    return CHUNK_LEFTOVER_DATA;
+    auto data_to_keep = last_data_in_working_buffer_.last(leftover_bytes_);
+    // Copy backwards, since if more than half of the data was unused then the
+    // source and destination will overlap.
+    std::copy_backward(data_to_keep.begin(), data_to_keep.end(),
+                       working_buffer_.begin());
   }
-
-  return CHUNK_READ_OKAY;
 }
 
 }  // namespace audio
diff --git a/src/audio/fatfs_audio_input.cpp b/src/audio/fatfs_audio_input.cpp
index 7f11805e..29e03784 100644
--- a/src/audio/fatfs_audio_input.cpp
+++ b/src/audio/fatfs_audio_input.cpp
@@ -1,5 +1,6 @@
 #include "fatfs_audio_input.hpp"
 
+#include <algorithm>
 #include <cstdint>
 #include <memory>
 #include <string>
@@ -10,33 +11,26 @@
 #include "audio_element.hpp"
 #include "chunk.hpp"
 #include "stream_buffer.hpp"
+#include "stream_event.hpp"
 #include "stream_message.hpp"
 
 static const char* kTag = "SRC";
 
 namespace audio {
 
-static const TickType_t kServiceInterval = pdMS_TO_TICKS(50);
-
-static const std::size_t kFileBufferSize = 1024 * 128;
-static const std::size_t kMinFileReadSize = 1024 * 4;
+// 32KiB to match the minimum himen region size.
+static const std::size_t kChunkSize = 1024;
 
 FatfsAudioInput::FatfsAudioInput(std::shared_ptr<drivers::SdStorage> storage)
     : IAudioElement(),
       storage_(storage),
-      raw_file_buffer_(static_cast<std::byte*>(
-          heap_caps_malloc(kFileBufferSize, MALLOC_CAP_SPIRAM))),
-      file_buffer_(raw_file_buffer_, kFileBufferSize),
-      file_buffer_read_pos_(file_buffer_.begin()),
-      file_buffer_write_pos_(file_buffer_.begin()),
       current_file_(),
-      is_file_open_(false),
-      chunk_writer_(nullptr) {
-  // TODO: create our chunk writer whenever the output buffer changes.
-}
+      is_file_open_(false) {}
+
+FatfsAudioInput::~FatfsAudioInput() {}
 
-FatfsAudioInput::~FatfsAudioInput() {
-  free(raw_file_buffer_);
+auto FatfsAudioInput::HasUnprocessedInput() -> bool {
+  return is_file_open_;
 }
 
 auto FatfsAudioInput::ProcessStreamInfo(const StreamInfo& info)
@@ -57,17 +51,12 @@ auto FatfsAudioInput::ProcessStreamInfo(const StreamInfo& info)
 
   is_file_open_ = true;
 
-  auto write_size =
-      WriteMessage(TYPE_STREAM_INFO,
-                   std::bind(&StreamInfo::Encode, info, std::placeholders::_1),
-                   output_buffer_->WriteBuffer());
+  std::unique_ptr<StreamInfo> new_info = std::make_unique<StreamInfo>(info);
+  new_info->ChunkSize(kChunkSize);
 
-  if (write_size.has_error()) {
-    return cpp::fail(IO_ERROR);
-  } else {
-    xMessageBufferSend(output_buffer_, output_buffer_->WriteBuffer().data(),
-                       write_size.value(), portMAX_DELAY);
-  }
+  auto event =
+      StreamEvent::CreateStreamInfo(input_events_, std::move(new_info));
+  SendOrBufferEvent(std::move(event));
 
   return {};
 }
@@ -77,110 +66,29 @@ auto FatfsAudioInput::ProcessChunk(const cpp::span<std::byte>& chunk)
   return cpp::fail(UNSUPPORTED_STREAM);
 }
 
-auto FatfsAudioInput::GetRingBufferDistance() const -> size_t {
-  if (file_buffer_read_pos_ == file_buffer_write_pos_) {
-    return 0;
-  }
-  if (file_buffer_read_pos_ < file_buffer_write_pos_) {
-    return file_buffer_write_pos_ - file_buffer_read_pos_;
-  }
-  return
-      // Read position to end of buffer.
-      (file_buffer_.end() - file_buffer_read_pos_)
-      // Start of buffer to write position.
-      + (file_buffer_write_pos_ - file_buffer_.begin());
-}
-
-auto FatfsAudioInput::ProcessIdle() -> cpp::result<void, AudioProcessingError> {
-  // First, see if we're able to fill up the input buffer with any more of the
-  // file's contents.
+auto FatfsAudioInput::Process() -> cpp::result<void, AudioProcessingError> {
   if (is_file_open_) {
-    size_t ringbuf_distance = GetRingBufferDistance();
-    if (file_buffer_.size() - ringbuf_distance > kMinFileReadSize) {
-      size_t read_size;
-      if (file_buffer_write_pos_ < file_buffer_read_pos_) {
-        // Don't worry about the start of buffer -> read pos size; we can get to
-        // it next iteration.
-        read_size = file_buffer_read_pos_ - file_buffer_write_pos_;
-      } else {
-        read_size = file_buffer_.begin() - file_buffer_write_pos_;
-      }
-
-      ESP_LOGI(kTag, "reading up to %d bytes", (int)read_size);
-
-      UINT bytes_read = 0;
-      FRESULT result =
-          f_read(&current_file_, std::addressof(file_buffer_write_pos_),
-                 read_size, &bytes_read);
-      if (result != FR_OK) {
-        ESP_LOGE(kTag, "file I/O error %d", result);
-        return cpp::fail(IO_ERROR);
-      }
-
-      ESP_LOGI(kTag, "actual read size %d bytes", (int)bytes_read);
-
-      if (f_eof(&current_file_)) {
-        f_close(&current_file_);
-        is_file_open_ = false;
-
-        // TODO: open the next file?
-      }
-
-      file_buffer_write_pos_ += bytes_read;
-      if (file_buffer_write_pos_ == file_buffer_.end()) {
-        file_buffer_write_pos_ = file_buffer_.begin();
-      }
+    auto dest_event = StreamEvent::CreateChunkData(input_events_, kChunkSize);
+    UINT bytes_read = 0;
+
+    FRESULT result =
+        f_read(&current_file_, dest_event->chunk_data.raw_bytes.get(),
+               kChunkSize, &bytes_read);
+    if (result != FR_OK) {
+      ESP_LOGE(kTag, "file I/O error %d", result);
+      return cpp::fail(IO_ERROR);
     }
-  } else if (GetRingBufferDistance() == 0) {
-    // We have no file open, and no data waiting to be written. We're out of
-    // stuff to do, so signal a pause.
-    return cpp::fail(OUT_OF_DATA);
-  }
 
-  // Now stream data into the output buffer until it's full.
-  while (GetRingBufferDistance() > 0) {
-    ESP_LOGI(kTag, "writing up to %d bytes", (int)GetRingBufferDistance());
-    ChunkWriteResult result = chunk_writer_->WriteChunkToStream(
-        [&](cpp::span<std::byte> d) { return SendChunk(d); }, kServiceInterval);
-
-    switch (result) {
-      case CHUNK_WRITE_OKAY:
-        break;
-      case CHUNK_WRITE_TIMEOUT:
-      case CHUNK_OUT_OF_DATA:
-        // Both of these are fine; we will pick back up where we left off in
-        // the next idle call.
-        return {};
-      default:
-        return cpp::fail(IO_ERROR);
+    dest_event->chunk_data.bytes =
+        dest_event->chunk_data.bytes.first(bytes_read);
+    SendOrBufferEvent(std::move(dest_event));
+
+    if (f_eof(&current_file_)) {
+      f_close(&current_file_);
+      is_file_open_ = false;
     }
   }
-
-  // We've finished writing out chunks, but there may be more of the file to
-  // read. Return, and begin again in the next idle call.
   return {};
 }
 
-auto FatfsAudioInput::SendChunk(cpp::span<std::byte> dest) -> size_t {
-  if (file_buffer_read_pos_ == file_buffer_write_pos_) {
-    return 0;
-  }
-  std::size_t chunk_size;
-  if (file_buffer_read_pos_ > file_buffer_write_pos_) {
-    chunk_size = file_buffer_.end() - file_buffer_read_pos_;
-  } else {
-    chunk_size = file_buffer_write_pos_ - file_buffer_read_pos_;
-  }
-  chunk_size = std::min(chunk_size, dest.size());
-
-  cpp::span<std::byte> source(file_buffer_read_pos_, chunk_size);
-  std::copy(source.begin(), source.end(), dest.begin());
-
-  file_buffer_read_pos_ = file_buffer_read_pos_ + chunk_size;
-  if (file_buffer_read_pos_ == file_buffer_.end()) {
-    file_buffer_read_pos_ = file_buffer_.begin();
-  }
-  return chunk_size;
-}
-
 }  // namespace audio
diff --git a/src/audio/i2s_audio_output.cpp b/src/audio/i2s_audio_output.cpp
index d853a06f..d041f726 100644
--- a/src/audio/i2s_audio_output.cpp
+++ b/src/audio/i2s_audio_output.cpp
@@ -92,22 +92,12 @@ auto I2SAudioOutput::ProcessChunk(const cpp::span<std::byte>& chunk)
   return dac_->WriteData(chunk, portMAX_DELAY);
 }
 
-auto I2SAudioOutput::IdleTimeout() const -> TickType_t {
-  return kIdleTimeBeforeMute;
-}
-
-auto I2SAudioOutput::ProcessIdle() -> cpp::result<void, AudioProcessingError> {
+auto I2SAudioOutput::Process() -> cpp::result<void, AudioProcessingError> {
   // TODO(jacqueline): Consider powering down the dac completely maybe?
   SetSoftMute(true);
   return {};
 }
 
-auto I2SAudioOutput::PrepareForPause() -> void {
-  // TODO(jacqueline): We ideally want to ensure we have enough samples in the
-  // DMA buffer here, so that soft mute can work properly.
-  SetSoftMute(true);
-}
-
 auto I2SAudioOutput::SetVolume(uint8_t volume) -> void {
   volume_ = volume;
   if (!is_soft_muted_) {
diff --git a/src/audio/include/audio_decoder.hpp b/src/audio/include/audio_decoder.hpp
index 0a2df76d..a2591d25 100644
--- a/src/audio/include/audio_decoder.hpp
+++ b/src/audio/include/audio_decoder.hpp
@@ -4,6 +4,7 @@
 #include <cstdint>
 #include <memory>
 
+#include "chunk.hpp"
 #include "ff.h"
 #include "span.hpp"
 
@@ -30,11 +31,13 @@ class AudioDecoder : public IAudioElement {
     return 1024;
   }
 
+  auto HasUnprocessedInput() -> bool override;
+
   auto ProcessStreamInfo(const StreamInfo& info)
       -> cpp::result<void, AudioProcessingError> override;
   auto ProcessChunk(const cpp::span<std::byte>& chunk)
       -> cpp::result<std::size_t, AudioProcessingError> override;
-  auto ProcessIdle() -> cpp::result<void, AudioProcessingError> override;
+  auto Process() -> cpp::result<void, AudioProcessingError> override;
 
   AudioDecoder(const AudioDecoder&) = delete;
   AudioDecoder& operator=(const AudioDecoder&) = delete;
@@ -42,8 +45,11 @@ class AudioDecoder : public IAudioElement {
  private:
   std::unique_ptr<codecs::ICodec> current_codec_;
   std::optional<StreamInfo> stream_info_;
+  std::optional<ChunkReader> chunk_reader_;
 
-  std::unique_ptr<ChunkWriter> chunk_writer_;
+  std::size_t chunk_size_;
+  bool has_samples_to_send_;
+  bool needs_more_input_;
 };
 
 }  // namespace audio
diff --git a/src/audio/include/audio_element.hpp b/src/audio/include/audio_element.hpp
index eb700180..8827a0c3 100644
--- a/src/audio/include/audio_element.hpp
+++ b/src/audio/include/audio_element.hpp
@@ -2,6 +2,8 @@
 
 #include <atomic>
 #include <cstdint>
+#include <deque>
+#include <memory>
 
 #include "freertos/FreeRTOS.h"
 
@@ -12,6 +14,7 @@
 #include "span.hpp"
 
 #include "stream_buffer.hpp"
+#include "stream_event.hpp"
 #include "stream_info.hpp"
 #include "types.hpp"
 
@@ -36,6 +39,8 @@ enum AudioProcessingError {
   OUT_OF_DATA,
 };
 
+static const size_t kEventQueueSize = 8;
+
 /*
  * One indepedentent part of an audio pipeline. Each element has an input and
  * output message stream, and is responsible for taking data from the input
@@ -51,38 +56,38 @@ enum AudioProcessingError {
  */
 class IAudioElement {
  public:
-  IAudioElement()
-      : input_buffer_(nullptr), output_buffer_(nullptr), state_(STATE_RUN) {}
-  virtual ~IAudioElement() {}
+  IAudioElement();
+  virtual ~IAudioElement();
 
   /*
    * Returns the stack size in bytes that this element requires. This should
    * be tuned according to the observed stack size of each element, as different
-   * elements have fairly different stack requirements.
+   * elements have fairly different stack requirements (particular decoders).
    */
   virtual auto StackSizeBytes() const -> std::size_t { return 2048; };
 
-  /*
-   * How long to wait for new data on the input stream before triggering a call
-   * to ProcessIdle(). If this is portMAX_DELAY (the default), then ProcessIdle
-   * will never be called.
-   */
-  virtual auto IdleTimeout() const -> TickType_t { return 10; }
-
   virtual auto InputMinChunkSize() const -> std::size_t { return 0; }
 
   /* Returns this element's input buffer. */
-  auto InputBuffer() const -> StreamBuffer* { return input_buffer_; }
+  auto InputEventQueue() const -> QueueHandle_t { return input_events_; }
 
   /* Returns this element's output buffer. */
-  auto OutputBuffer() const -> StreamBuffer* { return output_buffer_; }
+  auto OutputEventQueue() const -> QueueHandle_t { return output_events_; }
+
+  auto OutputEventQueue(const QueueHandle_t q) -> void { output_events_ = q; }
 
-  auto InputBuffer(StreamBuffer* b) -> void { input_buffer_ = b; }
+  auto HasUnflushedOutput() -> bool { return !buffered_output_.empty(); }
 
-  auto OutputBuffer(StreamBuffer* b) -> void { output_buffer_ = b; }
+  virtual auto HasUnprocessedInput() -> bool = 0;
 
-  auto ElementState() const -> ElementState { return state_; }
-  auto ElementState(enum ElementState e) -> void { state_ = e; }
+  auto IsOverBuffered() -> bool { return unprocessed_output_chunks_ > 4; }
+
+  auto FlushBufferedOutput() -> bool;
+
+  auto ElementState() const -> ElementState { return current_state_; }
+  auto ElementState(enum ElementState e) -> void { current_state_ = e; }
+
+  virtual auto OnChunkProcessed() -> void { unprocessed_output_chunks_--; }
 
   /*
    * Called when a StreamInfo message is received. Used to configure this
@@ -105,14 +110,26 @@ class IAudioElement {
    * time. This could be used to synthesize output, or to save memory by
    * releasing unused resources.
    */
-  virtual auto ProcessIdle() -> cpp::result<void, AudioProcessingError> = 0;
-
-  virtual auto PrepareForPause() -> void{};
+  virtual auto Process() -> cpp::result<void, AudioProcessingError> = 0;
 
  protected:
-  StreamBuffer* input_buffer_;
-  StreamBuffer* output_buffer_;
-  std::atomic<enum ElementState> state_;
+  auto SendOrBufferEvent(std::unique_ptr<StreamEvent> event) -> bool;
+
+  // Queue for events coming into this element. Owned by us.
+  QueueHandle_t input_events_;
+  // Queue for events going into the next element. Not owned by us, may be null
+  // if we're not yet in a pipeline.
+  // FIXME: it would be nicer if this was non-nullable.
+  QueueHandle_t output_events_;
+
+  // The number of output chunks that we have generated, but have not yet been
+  // processed by the next element in the pipeline. This includes any chunks
+  // that are currently help in buffered_output_.
+  int unprocessed_output_chunks_;
+  // Output events that have been generated, but are yet to be sent downstream.
+  std::deque<std::unique_ptr<StreamEvent>> buffered_output_;
+
+  enum ElementState current_state_;
 };
 
 }  // namespace audio
diff --git a/src/audio/include/audio_playback.hpp b/src/audio/include/audio_playback.hpp
index bffc3f02..f05ca327 100644
--- a/src/audio/include/audio_playback.hpp
+++ b/src/audio/include/audio_playback.hpp
@@ -43,11 +43,9 @@ class AudioPlayback {
 
  private:
   auto ConnectElements(IAudioElement* src, IAudioElement* sink) -> void;
-
-  StreamBuffer stream_start_;
-  StreamBuffer stream_end_;
-  std::vector<std::unique_ptr<StreamBuffer>> element_buffers_;
   std::vector<std::unique_ptr<AudioElementHandle>> element_handles_;
+
+  QueueHandle_t input_handle_;
 };
 
 }  // namespace audio
diff --git a/src/audio/include/chunk.hpp b/src/audio/include/chunk.hpp
index 5c7e73dd..6154ab25 100644
--- a/src/audio/include/chunk.hpp
+++ b/src/audio/include/chunk.hpp
@@ -2,6 +2,7 @@
 
 #include <cstddef>
 #include <cstdint>
+#include <memory>
 #include <optional>
 #include <string>
 
@@ -17,68 +18,12 @@
 
 namespace audio {
 
-enum ChunkWriteResult {
-  // Returned when the callback does not write any data.
-  CHUNK_WRITE_OKAY,
-  // Returned when the callback does not write any data.
-  CHUNK_OUT_OF_DATA,
-  // Returned when there is an error encoding a chunk header using cbor.
-  CHUNK_ENCODING_ERROR,
-  // Returned when max_wait expires without room in the stream buffer becoming
-  // available.
-  CHUNK_WRITE_TIMEOUT,
-};
-
-class ChunkWriter {
- public:
-  explicit ChunkWriter(StreamBuffer* buffer);
-  ~ChunkWriter();
-
-  auto Reset() -> void;
-
-  auto GetLastMessage() -> cpp::span<std::byte>;
-
-  /*
-   * Invokes the given callback to receive data, breaks the received data up
-   * into chunks with headers, and writes those chunks to the given output
-   * stream.
-   *
-   * The callback will be invoked with a byte buffer and its size. The callback
-   * should write as much data as it can to this buffer, and then return the
-   * number of bytes it wrote. Return a value of 0 to indicate that there is no
-   * more input to read.
-   */
-  auto WriteChunkToStream(std::function<size_t(cpp::span<std::byte>)> callback,
-                          TickType_t max_wait) -> ChunkWriteResult;
-
- private:
-  StreamBuffer* stream_;
-  std::size_t leftover_bytes_ = 0;
-};
-
-enum ChunkReadResult {
-  CHUNK_READ_OKAY,
-  // Returned when the chunk was read successfully, but the consumer did not
-  // use all of the data.
-  CHUNK_LEFTOVER_DATA,
-  // Returned an error in parsing the cbor-encoded header.
-  CHUNK_DECODING_ERROR,
-  // Returned when max_wait expired before any data was read.
-  CHUNK_READ_TIMEOUT,
-  // Returned when a non-chunk message is received.
-  CHUNK_STREAM_ENDED,
-  // Returned when the processing callback does not return a value.
-  CHUNK_PROCESSING_ERROR,
-};
-
 class ChunkReader {
  public:
-  explicit ChunkReader(StreamBuffer* buffer);
+  explicit ChunkReader(std::size_t chunk_size);
   ~ChunkReader();
 
-  auto Reset() -> void;
-
-  auto GetLastMessage() -> cpp::span<std::byte>;
+  auto HandleLeftovers(std::size_t bytes_used) -> void;
 
   /*
    * Reads chunks of data from the given input stream, and invokes the given
@@ -92,14 +37,13 @@ class ChunkReader {
    * If this function encounters a message in the stream that is not a chunk, it
    * will place the message at the start of the working_buffer and then return.
    */
-  auto ReadChunkFromStream(
-      std::function<std::optional<std::size_t>(cpp::span<std::byte>)> callback,
-      TickType_t max_wait) -> ChunkReadResult;
+  auto HandleNewData(cpp::span<std::byte> data) -> cpp::span<std::byte>;
 
  private:
-  StreamBuffer* stream_;
+  std::byte* raw_working_buffer_;
+  cpp::span<std::byte> working_buffer_;
+  cpp::span<std::byte> last_data_in_working_buffer_;
   std::size_t leftover_bytes_ = 0;
-  std::size_t last_message_size_ = 0;
 };
 
 }  // namespace audio
diff --git a/src/audio/include/fatfs_audio_input.hpp b/src/audio/include/fatfs_audio_input.hpp
index f3704f1d..5625d941 100644
--- a/src/audio/include/fatfs_audio_input.hpp
+++ b/src/audio/include/fatfs_audio_input.hpp
@@ -22,32 +22,22 @@ class FatfsAudioInput : public IAudioElement {
   explicit FatfsAudioInput(std::shared_ptr<drivers::SdStorage> storage);
   ~FatfsAudioInput();
 
+  auto HasUnprocessedInput() -> bool override;
+
   auto ProcessStreamInfo(const StreamInfo& info)
       -> cpp::result<void, AudioProcessingError> override;
   auto ProcessChunk(const cpp::span<std::byte>& chunk)
       -> cpp::result<std::size_t, AudioProcessingError> override;
-  auto ProcessIdle() -> cpp::result<void, AudioProcessingError> override;
-
-  auto SendChunk(cpp::span<std::byte> dest) -> size_t;
+  auto Process() -> cpp::result<void, AudioProcessingError> override;
 
   FatfsAudioInput(const FatfsAudioInput&) = delete;
   FatfsAudioInput& operator=(const FatfsAudioInput&) = delete;
 
  private:
-  auto GetRingBufferDistance() const -> size_t;
-
   std::shared_ptr<drivers::SdStorage> storage_;
 
-  std::byte* raw_file_buffer_;
-  cpp::span<std::byte> file_buffer_;
-  cpp::span<std::byte>::iterator file_buffer_read_pos_;
-  cpp::span<std::byte>::iterator pending_read_pos_;
-  cpp::span<std::byte>::iterator file_buffer_write_pos_;
-
   FIL current_file_;
   bool is_file_open_;
-
-  std::unique_ptr<ChunkWriter> chunk_writer_;
 };
 
 }  // namespace audio
diff --git a/src/audio/include/i2s_audio_output.hpp b/src/audio/include/i2s_audio_output.hpp
index 75a3be76..4fbcad49 100644
--- a/src/audio/include/i2s_audio_output.hpp
+++ b/src/audio/include/i2s_audio_output.hpp
@@ -22,19 +22,14 @@ class I2SAudioOutput : public IAudioElement {
                  std::unique_ptr<drivers::AudioDac> dac);
   ~I2SAudioOutput();
 
-  auto InputMinChunkSize() const -> std::size_t override {
-    // TODO(jacqueline): work out a good value here. Maybe similar to the total
-    // DMA buffer size?
-    return 128;
-  }
+  // TODO.
+  auto HasUnprocessedInput() -> bool override { return false; }
 
-  auto IdleTimeout() const -> TickType_t override;
   auto ProcessStreamInfo(const StreamInfo& info)
       -> cpp::result<void, AudioProcessingError> override;
   auto ProcessChunk(const cpp::span<std::byte>& chunk)
       -> cpp::result<std::size_t, AudioProcessingError> override;
-  auto ProcessIdle() -> cpp::result<void, AudioProcessingError> override;
-  auto PrepareForPause() -> void override;
+  auto Process() -> cpp::result<void, AudioProcessingError> override;
 
   I2SAudioOutput(const I2SAudioOutput&) = delete;
   I2SAudioOutput& operator=(const I2SAudioOutput&) = delete;
diff --git a/src/audio/include/stream_event.hpp b/src/audio/include/stream_event.hpp
new file mode 100644
index 00000000..4dfdab41
--- /dev/null
+++ b/src/audio/include/stream_event.hpp
@@ -0,0 +1,57 @@
+#pragma once
+
+#include <memory>
+
+#include "freertos/FreeRTOS.h"
+#include "freertos/queue.h"
+
+#include "span.hpp"
+#include "stream_info.hpp"
+
+namespace audio {
+
+struct StreamEvent {
+  static auto CreateStreamInfo(QueueHandle_t source,
+                               std::unique_ptr<StreamInfo> payload)
+      -> std::unique_ptr<StreamEvent>;
+  static auto CreateChunkData(QueueHandle_t source, std::size_t chunk_size)
+      -> std::unique_ptr<StreamEvent>;
+  static auto CreateChunkNotification(QueueHandle_t source)
+      -> std::unique_ptr<StreamEvent>;
+
+  StreamEvent();
+  ~StreamEvent();
+  StreamEvent(StreamEvent&&);
+
+  QueueHandle_t source;
+
+  enum {
+    UNINITIALISED,
+    STREAM_INFO,
+    CHUNK_DATA,
+    CHUNK_NOTIFICATION,
+  } tag;
+
+  union {
+    std::unique_ptr<StreamInfo> stream_info;
+
+    // Scott Meyers says:
+    // `About the only situation I can conceive of when a std::unique_ptr<T[]>
+    // would make sense would be when you’re using a C-like API that returns a
+    // raw pointer to a heap array that you assume ownership of.`
+    // :-)
+
+    struct {
+      std::unique_ptr<std::byte*> raw_bytes;
+      cpp::span<std::byte> bytes;
+    } chunk_data;
+
+    // FIXME: It would be nice to also support a pointer to himem data here, to
+    // save a little ordinary heap space.
+  };
+
+  StreamEvent(const StreamEvent&) = delete;
+  StreamEvent& operator=(const StreamEvent&) = delete;
+};
+
+}  // namespace audio
diff --git a/src/audio/include/stream_info.hpp b/src/audio/include/stream_info.hpp
index 45f10fc6..ed3096bb 100644
--- a/src/audio/include/stream_info.hpp
+++ b/src/audio/include/stream_info.hpp
@@ -7,6 +7,7 @@
 
 #include "cbor.h"
 #include "result.hpp"
+#include "sys/_stdint.h"
 
 namespace audio {
 
@@ -24,14 +25,24 @@ class StreamInfo {
 
   auto Channels() const -> const std::optional<uint8_t>& { return channels_; }
 
+  auto BitsPerSample(uint8_t bpp) -> void { bits_per_sample_ = bpp; }
+
   auto BitsPerSample() const -> const std::optional<uint8_t>& {
     return bits_per_sample_;
   }
 
+  auto SampleRate(uint16_t rate) -> void { sample_rate_ = rate; }
+
   auto SampleRate() const -> const std::optional<uint16_t>& {
     return sample_rate_;
   }
 
+  auto ChunkSize() const -> const std::optional<std::size_t>& {
+    return chunk_size_;
+  }
+
+  auto ChunkSize(std::size_t s) -> void { chunk_size_ = s; }
+
   auto Encode(CborEncoder& enc) -> std::optional<CborError>;
 
  private:
@@ -39,6 +50,7 @@ class StreamInfo {
   std::optional<uint8_t> channels_;
   std::optional<uint8_t> bits_per_sample_;
   std::optional<uint16_t> sample_rate_;
+  std::optional<size_t> chunk_size_;
 };
 
 }  // namespace audio
diff --git a/src/audio/stream_buffer.cpp b/src/audio/stream_buffer.cpp
index 9bdeaa72..71ab3965 100644
--- a/src/audio/stream_buffer.cpp
+++ b/src/audio/stream_buffer.cpp
@@ -17,10 +17,11 @@ StreamBuffer::StreamBuffer(std::size_t chunk_size, std::size_t buffer_size)
       raw_output_chunk_(static_cast<std::byte*>(
           heap_caps_malloc(chunk_size, MALLOC_CAP_SPIRAM))),
       output_chunk_(raw_output_chunk_, chunk_size) {
-        assert(input_chunk_.size() <= buffer_size);
-        assert(output_chunk_.size() <= buffer_size);
-        ESP_LOGI("streambuf", "created buffer of chunk size %d, total size %d", chunk_size, buffer_size);
-      }
+  assert(input_chunk_.size() <= buffer_size);
+  assert(output_chunk_.size() <= buffer_size);
+  ESP_LOGI("streambuf", "created buffer of chunk size %d, total size %d",
+           chunk_size, buffer_size);
+}
 
 StreamBuffer::~StreamBuffer() {
   vMessageBufferDelete(handle_);
diff --git a/src/audio/stream_event.cpp b/src/audio/stream_event.cpp
new file mode 100644
index 00000000..0a116297
--- /dev/null
+++ b/src/audio/stream_event.cpp
@@ -0,0 +1,75 @@
+#include "stream_event.hpp"
+#include <cstddef>
+#include <memory>
+
+namespace audio {
+
+auto StreamEvent::CreateStreamInfo(QueueHandle_t source,
+                                   std::unique_ptr<StreamInfo> payload)
+    -> std::unique_ptr<StreamEvent> {
+  auto event = std::make_unique<StreamEvent>();
+  event->tag = StreamEvent::STREAM_INFO;
+  event->source = source;
+  event->stream_info = std::move(payload);
+  return event;
+}
+
+auto StreamEvent::CreateChunkData(QueueHandle_t source, std::size_t chunk_size)
+    -> std::unique_ptr<StreamEvent> {
+  auto event = std::make_unique<StreamEvent>();
+  event->tag = StreamEvent::CHUNK_DATA;
+  event->source = source;
+
+  auto raw_bytes =
+      static_cast<std::byte*>(heap_caps_malloc(chunk_size, MALLOC_CAP_SPIRAM));
+
+  event->chunk_data.raw_bytes = std::make_unique<std::byte*>(raw_bytes);
+  event->chunk_data.bytes = cpp::span<std::byte>(raw_bytes, chunk_size);
+
+  return event;
+}
+
+auto StreamEvent::CreateChunkNotification(QueueHandle_t source)
+    -> std::unique_ptr<StreamEvent> {
+  auto event = std::make_unique<StreamEvent>();
+  event->tag = StreamEvent::CHUNK_NOTIFICATION;
+  event->source = source;
+  return event;
+}
+
+StreamEvent::StreamEvent() : tag(StreamEvent::UNINITIALISED) {}
+
+StreamEvent::~StreamEvent() {
+  switch (tag) {
+    case UNINITIALISED:
+      break;
+    case STREAM_INFO:
+      stream_info.reset();
+      break;
+    case CHUNK_DATA:
+      chunk_data.raw_bytes.reset();
+      break;
+    case CHUNK_NOTIFICATION:
+      break;
+  }
+}
+
+StreamEvent::StreamEvent(StreamEvent&& other) {
+  tag = other.tag;
+  source = other.source;
+  switch (tag) {
+    case UNINITIALISED:
+      break;
+    case STREAM_INFO:
+      stream_info = std::move(other.stream_info);
+      break;
+    case CHUNK_DATA:
+      chunk_data = std::move(other.chunk_data);
+      break;
+    case CHUNK_NOTIFICATION:
+      break;
+  }
+  other.tag = StreamEvent::UNINITIALISED;
+}
+
+}  // namespace audio
diff --git a/src/audio/stream_message.cpp b/src/audio/stream_message.cpp
index 055e7e96..0f3b56e9 100644
--- a/src/audio/stream_message.cpp
+++ b/src/audio/stream_message.cpp
@@ -3,7 +3,6 @@
 #include <cstdint>
 
 #include "cbor.h"
-#include "esp-idf/components/cbor/tinycbor/src/cbor.h"
 #include "span.hpp"
 
 namespace audio {