Count samples going in and out of the drain buffer

This is a more accurate way of knowing which track is playing when, and
also simplifies a lot of fragile logic in audio_fsm

18 files changed, 332 insertions, 263 deletions

diff --git a/src/drivers/bluetooth.cpp b/src/drivers/bluetooth.cpp
index 4caffae7..fcb764f6 100644
--- a/src/drivers/bluetooth.cpp
+++ b/src/drivers/bluetooth.cpp
@@ -38,6 +38,7 @@ namespace drivers {
 
 DRAM_ATTR static StreamBufferHandle_t sStream = nullptr;
 DRAM_ATTR static std::atomic<float> sVolumeFactor = 1.f;
+DRAM_ATTR static std::atomic<uint32_t> sSamplesUsed = 0;
 
 static tasks::WorkerPool* sBgWorker;
 
@@ -47,8 +48,8 @@ auto gap_cb(esp_bt_gap_cb_event_t event, esp_bt_gap_cb_param_t* param) -> void {
       bluetooth::events::internal::Gap{.type = event, .param = param});
 }
 
-auto avrcp_cb(esp_avrc_ct_cb_event_t event,
-              esp_avrc_ct_cb_param_t* param) -> void {
+auto avrcp_cb(esp_avrc_ct_cb_event_t event, esp_avrc_ct_cb_param_t* param)
+    -> void {
   esp_avrc_ct_cb_param_t copy = *param;
   sBgWorker->Dispatch<void>([=]() {
     auto lock = bluetooth::BluetoothState::lock();
@@ -73,6 +74,13 @@ IRAM_ATTR auto a2dp_data_cb(uint8_t* buf, int32_t buf_size) -> int32_t {
   }
   size_t bytes_received = xStreamBufferReceive(stream, buf, buf_size, 0);
 
+  size_t samples_received = bytes_received / 2;
+  if (UINT32_MAX - sSamplesUsed < samples_received) {
+    sSamplesUsed = samples_received - (UINT32_MAX - sSamplesUsed);
+  } else {
+    sSamplesUsed += samples_received;
+  }
+
   // Apply software volume scaling.
   int16_t* samples = reinterpret_cast<int16_t*>(buf);
   float factor = sVolumeFactor.load();
@@ -165,6 +173,10 @@ auto Bluetooth::SetVolumeFactor(float f) -> void {
   sVolumeFactor = f;
 }
 
+auto Bluetooth::SamplesUsed() -> uint32_t {
+  return sSamplesUsed;
+}
+
 auto Bluetooth::SetEventHandler(std::function<void(bluetooth::Event)> cb)
     -> void {
   auto lock = bluetooth::BluetoothState::lock();
diff --git a/src/drivers/i2s_dac.cpp b/src/drivers/i2s_dac.cpp
index a271fce0..e5efe198 100644
--- a/src/drivers/i2s_dac.cpp
+++ b/src/drivers/i2s_dac.cpp
@@ -5,6 +5,7 @@
  */
 
 #include "drivers/i2s_dac.hpp"
+#include <stdint.h>
 
 #include <cmath>
 #include <cstdint>
@@ -140,11 +141,10 @@ auto I2SDac::SetPaused(bool paused) -> void {
   }
 }
 
-static volatile bool sSwapWords = false;
+DRAM_ATTR static volatile bool sSwapWords = false;
 
-auto I2SDac::Reconfigure(Channels ch,
-                         BitsPerSample bps,
-                         SampleRate rate) -> void {
+auto I2SDac::Reconfigure(Channels ch, BitsPerSample bps, SampleRate rate)
+    -> void {
   std::lock_guard<std::mutex> lock(configure_mutex_);
 
   if (i2s_active_) {
@@ -217,6 +217,8 @@ auto I2SDac::WriteData(const std::span<const std::byte>& data) -> void {
   }
 }
 
+DRAM_ATTR static volatile uint32_t sSamplesRead = 0;
+
 extern "C" IRAM_ATTR auto callback(i2s_chan_handle_t handle,
                                    i2s_event_data_t* event,
                                    void* user_ctx) -> bool {
@@ -235,6 +237,14 @@ extern "C" IRAM_ATTR auto callback(i2s_chan_handle_t handle,
   size_t bytes_written =
       xStreamBufferReceiveFromISR(src, buf, event->size, &ret);
 
+  // Assume 16 bit samples.
+  size_t samples = bytes_written / 2;
+  if (UINT32_MAX - sSamplesRead < samples) {
+    sSamplesRead = samples - (UINT32_MAX - sSamplesRead);
+  } else {
+    sSamplesRead = sSamplesRead + samples;
+  }
+
   // The ESP32's I2S peripheral has a different endianness to its processors.
   // ESP-IDF handles this difference for stereo channels, but not for mono
   // channels. We therefore sometimes need to swap each pair of words as they're
@@ -276,6 +286,10 @@ auto I2SDac::SetSource(StreamBufferHandle_t buffer) -> void {
   }
 }
 
+auto I2SDac::SamplesUsed() -> uint32_t {
+  return sSamplesRead;
+}
+
 auto I2SDac::set_channel(bool enabled) -> void {
   if (i2s_active_ == enabled) {
     return;
diff --git a/src/drivers/include/drivers/bluetooth.hpp b/src/drivers/include/drivers/bluetooth.hpp
index 5960de7e..ad61fcc1 100644
--- a/src/drivers/include/drivers/bluetooth.hpp
+++ b/src/drivers/include/drivers/bluetooth.hpp
@@ -13,10 +13,10 @@
 #include <freertos/stream_buffer.h>
 #include <stdint.h>
 #include "drivers/bluetooth_types.hpp"
+#include "drivers/nvs.hpp"
 #include "esp_a2dp_api.h"
 #include "esp_avrc_api.h"
 #include "esp_gap_bt_api.h"
-#include "drivers/nvs.hpp"
 #include "tasks.hpp"
 #include "tinyfsm.hpp"
 #include "tinyfsm/include/tinyfsm.hpp"
@@ -44,6 +44,7 @@ class Bluetooth {
 
   auto SetSource(StreamBufferHandle_t) -> void;
   auto SetVolumeFactor(float) -> void;
+  auto SamplesUsed() -> uint32_t;
 
   auto SetEventHandler(std::function<void(bluetooth::Event)> cb) -> void;
 };
diff --git a/src/drivers/include/drivers/i2s_dac.hpp b/src/drivers/include/drivers/i2s_dac.hpp
index 5e81f875..0776dbab 100644
--- a/src/drivers/include/drivers/i2s_dac.hpp
+++ b/src/drivers/include/drivers/i2s_dac.hpp
@@ -11,8 +11,8 @@
 #include <functional>
 #include <memory>
 #include <optional>
-#include <utility>
 #include <span>
+#include <utility>
 
 #include "driver/i2s_std.h"
 #include "driver/i2s_types.h"
@@ -71,6 +71,8 @@ class I2SDac {
   auto WriteData(const std::span<const std::byte>& data) -> void;
   auto SetSource(StreamBufferHandle_t buffer) -> void;
 
+  auto SamplesUsed() -> uint32_t;
+
   // Not copyable or movable.
   I2SDac(const I2SDac&) = delete;
   I2SDac& operator=(const I2SDac&) = delete;
diff --git a/src/tangara/audio/audio_decoder.cpp b/src/tangara/audio/audio_decoder.cpp
index 0e38bca8..ee06d984 100644
--- a/src/tangara/audio/audio_decoder.cpp
+++ b/src/tangara/audio/audio_decoder.cpp
@@ -92,7 +92,7 @@ void Decoder::Main() {
       // If we were already decoding, then make sure we finish up the current
       // file gracefully.
       if (stream_) {
-        finishDecode();
+        finishDecode(true);
       }
 
       // Ensure there's actually stream data; we might have been given nullptr
@@ -106,7 +106,7 @@ void Decoder::Main() {
     }
 
     if (!continueDecode()) {
-      finishDecode();
+      finishDecode(false);
     }
   }
 }
@@ -179,12 +179,16 @@ auto Decoder::continueDecode() -> bool {
   return !res->is_stream_finished;
 }
 
-auto Decoder::finishDecode() -> void {
+auto Decoder::finishDecode(bool cancel) -> void {
   assert(track_);
 
   // Tell everyone we're finished.
-  events::Audio().Dispatch(internal::DecodingFinished{.track = track_});
-  processor_->endStream();
+  if (cancel) {
+    events::Audio().Dispatch(internal::DecodingCancelled{.track = track_});
+  } else {
+    events::Audio().Dispatch(internal::DecodingFinished{.track = track_});
+  }
+  processor_->endStream(cancel);
 
   // Clean up after ourselves.
   stream_.reset();
diff --git a/src/tangara/audio/audio_decoder.hpp b/src/tangara/audio/audio_decoder.hpp
index ee68290c..64561d9d 100644
--- a/src/tangara/audio/audio_decoder.hpp
+++ b/src/tangara/audio/audio_decoder.hpp
@@ -40,7 +40,7 @@ class Decoder {
 
   auto prepareDecode(std::shared_ptr<TaggedStream>) -> void;
   auto continueDecode() -> bool;
-  auto finishDecode() -> void;
+  auto finishDecode(bool cancel) -> void;
 
   std::shared_ptr<SampleProcessor> processor_;
 
diff --git a/src/tangara/audio/audio_events.hpp b/src/tangara/audio/audio_events.hpp
index bb8bf834..f7eaba67 100644
--- a/src/tangara/audio/audio_events.hpp
+++ b/src/tangara/audio/audio_events.hpp
@@ -84,13 +84,6 @@ struct PlaybackUpdate : tinyfsm::Event {
 struct SetTrack : tinyfsm::Event {
   std::variant<std::string, database::TrackId, std::monostate> new_track;
   std::optional<uint32_t> seek_to_second;
-
-  enum Transition {
-    kHardCut,
-    kGapless,
-    // TODO: kCrossFade
-  };
-  Transition transition;
 };
 
 struct TogglePlayPause : tinyfsm::Event {
@@ -146,21 +139,25 @@ struct DecodingFailedToStart : tinyfsm::Event {
   std::shared_ptr<TrackInfo> track;
 };
 
+struct DecodingCancelled : tinyfsm::Event {
+  std::shared_ptr<TrackInfo> track;
+};
+
 struct DecodingFinished : tinyfsm::Event {
   std::shared_ptr<TrackInfo> track;
 };
 
 struct StreamStarted : tinyfsm::Event {
   std::shared_ptr<TrackInfo> track;
-  IAudioOutput::Format src_format;
-  IAudioOutput::Format dst_format;
+  IAudioOutput::Format sink_format;
+  uint32_t cue_at_sample;
 };
 
-struct StreamUpdate : tinyfsm::Event {
-  uint32_t samples_sunk;
+struct StreamEnded : tinyfsm::Event {
+  uint32_t cue_at_sample;
 };
 
-struct StreamEnded : tinyfsm::Event {};
+struct StreamHeartbeat : tinyfsm::Event {};
 
 }  // namespace internal
 
diff --git a/src/tangara/audio/audio_fsm.cpp b/src/tangara/audio/audio_fsm.cpp
index d437cdbf..a7c006f5 100644
--- a/src/tangara/audio/audio_fsm.cpp
+++ b/src/tangara/audio/audio_fsm.cpp
@@ -5,8 +5,8 @@
  */
 
 #include "audio/audio_fsm.hpp"
-#include <stdint.h>
 
+#include <cstdint>
 #include <future>
 #include <memory>
 #include <variant>
@@ -18,6 +18,7 @@
 #include "freertos/FreeRTOS.h"
 #include "freertos/portmacro.h"
 #include "freertos/projdefs.h"
+#include "tinyfsm.hpp"
 
 #include "audio/audio_decoder.hpp"
 #include "audio/audio_events.hpp"
@@ -25,6 +26,7 @@
 #include "audio/bt_audio_output.hpp"
 #include "audio/fatfs_stream_factory.hpp"
 #include "audio/i2s_audio_output.hpp"
+#include "audio/stream_cues.hpp"
 #include "audio/track_queue.hpp"
 #include "database/future_fetcher.hpp"
 #include "database/track.hpp"
@@ -38,7 +40,6 @@
 #include "sample.hpp"
 #include "system_fsm/service_locator.hpp"
 #include "system_fsm/system_events.hpp"
-#include "tinyfsm.hpp"
 
 namespace audio {
 
@@ -47,11 +48,13 @@ namespace audio {
 std::shared_ptr<system_fsm::ServiceLocator> AudioState::sServices;
 
 std::shared_ptr<FatfsStreamFactory> AudioState::sStreamFactory;
+
 std::unique_ptr<Decoder> AudioState::sDecoder;
 std::shared_ptr<SampleProcessor> AudioState::sSampleProcessor;
+
+std::shared_ptr<IAudioOutput> AudioState::sOutput;
 std::shared_ptr<I2SAudioOutput> AudioState::sI2SOutput;
 std::shared_ptr<BluetoothAudioOutput> AudioState::sBtOutput;
-std::shared_ptr<IAudioOutput> AudioState::sOutput;
 
 // Two seconds of samples for two channels, at a representative sample rate.
 constexpr size_t kDrainLatencySamples = 48000 * 2 * 2;
@@ -61,30 +64,33 @@ constexpr size_t kDrainBufferSize =
 StreamBufferHandle_t AudioState::sDrainBuffer;
 std::optional<IAudioOutput::Format> AudioState::sDrainFormat;
 
-std::shared_ptr<TrackInfo> AudioState::sCurrentTrack;
-uint64_t AudioState::sCurrentSamples;
-bool AudioState::sCurrentTrackIsFromQueue;
+StreamCues AudioState::sStreamCues;
 
-std::shared_ptr<TrackInfo> AudioState::sNextTrack;
-uint64_t AudioState::sNextTrackCueSamples;
-bool AudioState::sNextTrackIsFromQueue;
-
-bool AudioState::sIsResampling;
 bool AudioState::sIsPaused = true;
 
-auto AudioState::currentPositionSeconds() -> std::optional<uint32_t> {
-  if (!sCurrentTrack || !sDrainFormat) {
-    return {};
+auto AudioState::emitPlaybackUpdate(bool paused) -> void {
+  std::optional<uint32_t> position;
+  auto current = sStreamCues.current();
+  if (current.first && sDrainFormat) {
+    position = (current.second /
+                (sDrainFormat->num_channels * sDrainFormat->sample_rate)) +
+               current.first->start_offset.value_or(0);
   }
-  return sCurrentSamples /
-         (sDrainFormat->num_channels * sDrainFormat->sample_rate);
+
+  PlaybackUpdate event{
+      .current_track = current.first,
+      .track_position = position,
+      .paused = paused,
+  };
+
+  events::System().Dispatch(event);
+  events::Ui().Dispatch(event);
 }
 
 void AudioState::react(const QueueUpdate& ev) {
   SetTrack cmd{
       .new_track = std::monostate{},
       .seek_to_second = {},
-      .transition = SetTrack::Transition::kHardCut,
   };
 
   auto current = sServices->track_queue().current();
@@ -97,20 +103,13 @@ void AudioState::react(const QueueUpdate& ev) {
       if (!ev.current_changed) {
         return;
       }
-      sNextTrackIsFromQueue = true;
-      cmd.transition = SetTrack::Transition::kHardCut;
       break;
     case QueueUpdate::kRepeatingLastTrack:
-      sNextTrackIsFromQueue = true;
-      cmd.transition = SetTrack::Transition::kGapless;
       break;
     case QueueUpdate::kTrackFinished:
       if (!ev.current_changed) {
         cmd.new_track = std::monostate{};
-      } else {
-        sNextTrackIsFromQueue = true;
       }
-      cmd.transition = SetTrack::Transition::kGapless;
       break;
     case QueueUpdate::kDeserialised:
     default:
@@ -123,32 +122,9 @@ void AudioState::react(const QueueUpdate& ev) {
 }
 
 void AudioState::react(const SetTrack& ev) {
-  // Remember the current track if there is one, since we need to preserve some
-  // of the state if it turns out this SetTrack event corresponds to seeking
-  // within the current track.
-  std::string prev_uri;
-  bool prev_from_queue = false;
-  if (sCurrentTrack) {
-    prev_uri = sCurrentTrack->uri;
-    prev_from_queue = sCurrentTrackIsFromQueue;
-  }
-
-  if (ev.transition == SetTrack::Transition::kHardCut) {
-    sCurrentTrack.reset();
-    sCurrentSamples = 0;
-    sCurrentTrackIsFromQueue = false;
-    clearDrainBuffer();
-  }
-
   if (std::holds_alternative<std::monostate>(ev.new_track)) {
     ESP_LOGI(kTag, "playback finished, awaiting drain");
     sDecoder->open({});
-    awaitEmptyDrainBuffer();
-    sCurrentTrack.reset();
-    sDrainFormat.reset();
-    sCurrentSamples = 0;
-    sCurrentTrackIsFromQueue = false;
-    transit<states::Standby>();
     return;
   }
 
@@ -166,81 +142,62 @@ void AudioState::react(const SetTrack& ev) {
           sStreamFactory->create(std::get<std::string>(new_track), seek_to);
     }
 
-    // This was a seek or replay within the same track; don't forget where
-    // the track originally came from.
-    // FIXME:
-    // sNextTrackIsFromQueue = prev_from_queue;
     sDecoder->open(stream);
   });
 }
 
 void AudioState::react(const TogglePlayPause& ev) {
   sIsPaused = !ev.set_to.value_or(sIsPaused);
-  if (!sIsPaused && is_in_state<states::Standby>() && sCurrentTrack) {
+  if (!sIsPaused && is_in_state<states::Standby>() &&
+      sStreamCues.current().first) {
     transit<states::Playback>();
   } else if (sIsPaused && is_in_state<states::Playback>()) {
     transit<states::Standby>();
   }
 }
 
-void AudioState::react(const internal::StreamStarted& ev) {
-  sDrainFormat = ev.dst_format;
-  sIsResampling = ev.src_format != ev.dst_format;
-
-  sNextTrack = ev.track;
-  sNextTrackCueSamples = sCurrentSamples + (kDrainLatencySamples / 2);
-
-  ESP_LOGI(kTag, "new stream %s %u ch @ %lu hz (resample=%i)",
-           ev.track->uri.c_str(), sDrainFormat->num_channels,
-           sDrainFormat->sample_rate, sIsResampling);
-}
-
-void AudioState::react(const internal::StreamEnded&) {
-  ESP_LOGI(kTag, "stream ended");
-
-  if (sCurrentTrackIsFromQueue) {
-    sServices->track_queue().finish();
-  } else {
-    tinyfsm::FsmList<AudioState>::dispatch(SetTrack{
-        .new_track = std::monostate{},
-        .seek_to_second = {},
-        .transition = SetTrack::Transition::kGapless,
-    });
-  }
+void AudioState::react(const internal::DecodingFinished& ev) {
+  // If we just finished playing whatever's at the front of the queue, then we
+  // need to advanve and start playing the next one ASAP in order to continue
+  // gaplessly.
+  sServices->bg_worker().Dispatch<void>([=]() {
+    auto& queue = sServices->track_queue();
+    auto current = queue.current();
+    if (!current) {
+      return;
+    }
+    auto db = sServices->database().lock();
+    if (!db) {
+      return;
+    }
+    auto path = db->getTrackPath(*current);
+    if (!path) {
+      return;
+    }
+    if (*path == ev.track->uri) {
+      queue.finish();
+    }
+  });
 }
 
-void AudioState::react(const internal::StreamUpdate& ev) {
-  sCurrentSamples += ev.samples_sunk;
-
-  if (sNextTrack && sCurrentSamples >= sNextTrackCueSamples) {
-    ESP_LOGI(kTag, "next track is now sinking");
-    sCurrentTrack = sNextTrack;
-    sCurrentSamples -= sNextTrackCueSamples;
-    sCurrentSamples += sNextTrack->start_offset.value_or(0) *
-                       (sDrainFormat->num_channels * sDrainFormat->sample_rate);
-    sCurrentTrackIsFromQueue = sNextTrackIsFromQueue;
-
-    sNextTrack.reset();
-    sNextTrackCueSamples = 0;
-    sNextTrackIsFromQueue = false;
+void AudioState::react(const internal::StreamStarted& ev) {
+  if (sDrainFormat != ev.sink_format) {
+    sDrainFormat = ev.sink_format;
+    ESP_LOGI(kTag, "sink_format=%u ch @ %lu hz", sDrainFormat->num_channels,
+             sDrainFormat->sample_rate);
   }
 
-  if (sCurrentTrack) {
-    PlaybackUpdate event{
-        .current_track = sCurrentTrack,
-        .track_position = currentPositionSeconds(),
-        .paused = !is_in_state<states::Playback>(),
-    };
-    events::System().Dispatch(event);
-    events::Ui().Dispatch(event);
-  }
+  sStreamCues.addCue(ev.track, ev.cue_at_sample);
 
-  if (sCurrentTrack && !sIsPaused && !is_in_state<states::Playback>()) {
-    ESP_LOGI(kTag, "ready to play!");
+  if (!sIsPaused && !is_in_state<states::Playback>()) {
     transit<states::Playback>();
   }
 }
 
+void AudioState::react(const internal::StreamEnded& ev) {
+  sStreamCues.addCue({}, ev.cue_at_sample);
+}
+
 void AudioState::react(const system_fsm::BluetoothEvent& ev) {
   if (ev.event != drivers::bluetooth::Event::kConnectionStateChanged) {
     return;
@@ -276,14 +233,6 @@ void AudioState::react(const StepDownVolume& ev) {
   }
 }
 
-void AudioState::react(const system_fsm::HasPhonesChanged& ev) {
-  if (ev.has_headphones) {
-    ESP_LOGI(kTag, "headphones in!");
-  } else {
-    ESP_LOGI(kTag, "headphones out!");
-  }
-}
-
 void AudioState::react(const SetVolume& ev) {
   if (ev.db.has_value()) {
     if (sOutput->SetVolumeDb(ev.db.value())) {
@@ -354,43 +303,6 @@ void AudioState::react(const OutputModeChanged& ev) {
   }
 }
 
-auto AudioState::clearDrainBuffer() -> void {
-  // Tell the decoder to stop adding new samples. This might not take effect
-  // immediately, since the decoder might currently be stuck waiting for space
-  // to become available in the drain buffer.
-  sDecoder->open({});
-
-  auto mode = sOutput->mode();
-  if (mode == IAudioOutput::Modes::kOnPlaying) {
-    // If we're currently playing, then the drain buffer will be actively
-    // draining on its own. Just keep trying to reset until it works.
-    while (xStreamBufferReset(sDrainBuffer) != pdPASS) {
-    }
-  } else {
-    // If we're not currently playing, then we need to actively pull samples
-    // out of the drain buffer to unblock the decoder.
-    while (!xStreamBufferIsEmpty(sDrainBuffer)) {
-      // Read a little to unblock the decoder.
-      uint8_t drain[2048];
-      xStreamBufferReceive(sDrainBuffer, drain, sizeof(drain), 0);
-
-      // Try to quickly discard the rest.
-      xStreamBufferReset(sDrainBuffer);
-    }
-  }
-}
-
-auto AudioState::awaitEmptyDrainBuffer() -> void {
-  if (is_in_state<states::Playback>()) {
-    for (int i = 0; i < 10 && !xStreamBufferIsEmpty(sDrainBuffer); i++) {
-      vTaskDelay(pdMS_TO_TICKS(250));
-    }
-  }
-  if (!xStreamBufferIsEmpty(sDrainBuffer)) {
-    clearDrainBuffer();
-  }
-}
-
 auto AudioState::commitVolume() -> void {
   auto mode = sServices->nvs().OutputMode();
   auto vol = sOutput->GetVolume();
@@ -455,7 +367,7 @@ void Uninitialised::react(const system_fsm::BootComplete& ev) {
       .left_bias = nvs.AmpLeftBias(),
   });
 
-  sSampleProcessor.reset(new SampleProcessor());
+  sSampleProcessor.reset(new SampleProcessor(sDrainBuffer));
   sSampleProcessor->SetOutput(sOutput);
 
   sDecoder.reset(Decoder::Start(sSampleProcessor));
@@ -470,7 +382,8 @@ void Standby::react(const system_fsm::KeyLockChanged& ev) {
   if (!ev.locking) {
     return;
   }
-  sServices->bg_worker().Dispatch<void>([this]() {
+  auto current = sStreamCues.current();
+  sServices->bg_worker().Dispatch<void>([=]() {
     auto db = sServices->database().lock();
     if (!db) {
       return;
@@ -483,10 +396,13 @@ void Standby::react(const system_fsm::KeyLockChanged& ev) {
     }
     db->put(kQueueKey, queue.serialise());
 
-    if (sCurrentTrack) {
+    if (current.first && sDrainFormat) {
+      uint32_t seconds = (current.second / (sDrainFormat->num_channels *
+                                            sDrainFormat->sample_rate)) +
+                         current.first->start_offset.value_or(0);
       cppbor::Array current_track{
-          cppbor::Tstr{sCurrentTrack->uri},
-          cppbor::Uint{currentPositionSeconds().value_or(0)},
+          cppbor::Tstr{current.first->uri},
+          cppbor::Uint{seconds},
       };
       db->put(kCurrentFileKey, current_track.toString());
     }
@@ -521,7 +437,6 @@ void Standby::react(const system_fsm::SdStateChanged& ev) {
         events::Audio().Dispatch(SetTrack{
             .new_track = filename,
             .seek_to_second = pos,
-            .transition = SetTrack::Transition::kHardCut,
         });
       }
     }
@@ -537,32 +452,29 @@ void Standby::react(const system_fsm::SdStateChanged& ev) {
   });
 }
 
+static TimerHandle_t sHeartbeatTimer;
+
+static void heartbeat(TimerHandle_t) {
+  events::Audio().Dispatch(internal::StreamHeartbeat{});
+}
+
 void Playback::entry() {
   ESP_LOGI(kTag, "audio output resumed");
   sOutput->mode(IAudioOutput::Modes::kOnPlaying);
+  emitPlaybackUpdate(false);
 
-  PlaybackUpdate event{
-      .current_track = sCurrentTrack,
-      .track_position = currentPositionSeconds(),
-      .paused = false,
-  };
-
-  events::System().Dispatch(event);
-  events::Ui().Dispatch(event);
+  if (!sHeartbeatTimer) {
+    sHeartbeatTimer =
+        xTimerCreate("stream", pdMS_TO_TICKS(250), true, NULL, heartbeat);
+  }
+  xTimerStart(sHeartbeatTimer, portMAX_DELAY);
 }
 
 void Playback::exit() {
   ESP_LOGI(kTag, "audio output paused");
+  xTimerStop(sHeartbeatTimer, portMAX_DELAY);
   sOutput->mode(IAudioOutput::Modes::kOnPaused);
-
-  PlaybackUpdate event{
-      .current_track = sCurrentTrack,
-      .track_position = currentPositionSeconds(),
-      .paused = true,
-  };
-
-  events::System().Dispatch(event);
-  events::Ui().Dispatch(event);
+  emitPlaybackUpdate(true);
 }
 
 void Playback::react(const system_fsm::SdStateChanged& ev) {
@@ -571,6 +483,17 @@ void Playback::react(const system_fsm::SdStateChanged& ev) {
   }
 }
 
+void Playback::react(const internal::StreamHeartbeat& ev) {
+  sStreamCues.update(sOutput->samplesUsed());
+  auto current = sStreamCues.current();
+
+  if (!current.first) {
+    transit<Standby>();
+  } else {
+    emitPlaybackUpdate(false);
+  }
+}
+
 }  // namespace states
 
 }  // namespace audio
diff --git a/src/tangara/audio/audio_fsm.hpp b/src/tangara/audio/audio_fsm.hpp
index 90e71404..03aaddcb 100644
--- a/src/tangara/audio/audio_fsm.hpp
+++ b/src/tangara/audio/audio_fsm.hpp
@@ -11,6 +11,7 @@
 #include <memory>
 #include <vector>
 
+#include "audio/stream_cues.hpp"
 #include "tinyfsm.hpp"
 
 #include "audio/audio_decoder.hpp"
@@ -46,13 +47,13 @@ class AudioState : public tinyfsm::Fsm<AudioState> {
   void react(const SetTrack&);
   void react(const TogglePlayPause&);
 
+  void react(const internal::DecodingFinished&);
   void react(const internal::StreamStarted&);
-  void react(const internal::StreamUpdate&);
   void react(const internal::StreamEnded&);
+  virtual void react(const internal::StreamHeartbeat&) {}
 
   void react(const StepUpVolume&);
   void react(const StepDownVolume&);
-  virtual void react(const system_fsm::HasPhonesChanged&);
 
   void react(const SetVolume&);
   void react(const SetVolumeLimit&);
@@ -66,10 +67,7 @@ class AudioState : public tinyfsm::Fsm<AudioState> {
   virtual void react(const system_fsm::BluetoothEvent&);
 
  protected:
-  auto clearDrainBuffer() -> void;
-  auto awaitEmptyDrainBuffer() -> void;
-
-  auto playTrack(database::TrackId id) -> void;
+  auto emitPlaybackUpdate(bool paused) -> void;
   auto commitVolume() -> void;
 
   static std::shared_ptr<system_fsm::ServiceLocator> sServices;
@@ -83,19 +81,10 @@ class AudioState : public tinyfsm::Fsm<AudioState> {
 
   static StreamBufferHandle_t sDrainBuffer;
 
-  static std::shared_ptr<TrackInfo> sCurrentTrack;
-  static uint64_t sCurrentSamples;
+  static StreamCues sStreamCues;
   static std::optional<IAudioOutput::Format> sDrainFormat;
-  static bool sCurrentTrackIsFromQueue;
-
-  static std::shared_ptr<TrackInfo> sNextTrack;
-  static uint64_t sNextTrackCueSamples;
-  static bool sNextTrackIsFromQueue;
 
-  static bool sIsResampling;
   static bool sIsPaused;
-
-  auto currentPositionSeconds() -> std::optional<uint32_t>;
 };
 
 namespace states {
@@ -122,6 +111,7 @@ class Playback : public AudioState {
   void exit() override;
 
   void react(const system_fsm::SdStateChanged&) override;
+  void react(const internal::StreamHeartbeat&) override;
 
   using AudioState::react;
 };
diff --git a/src/tangara/audio/audio_sink.hpp b/src/tangara/audio/audio_sink.hpp
index f31d0d75..0b133a8d 100644
--- a/src/tangara/audio/audio_sink.hpp
+++ b/src/tangara/audio/audio_sink.hpp
@@ -75,6 +75,7 @@ class IAudioOutput {
 
   virtual auto PrepareFormat(const Format&) -> Format = 0;
   virtual auto Configure(const Format& format) -> void = 0;
+  virtual auto samplesUsed() -> uint32_t = 0;
 
   auto stream() -> StreamBufferHandle_t { return stream_; }
 
diff --git a/src/tangara/audio/bt_audio_output.cpp b/src/tangara/audio/bt_audio_output.cpp
index 637c0c9a..f1b4c26c 100644
--- a/src/tangara/audio/bt_audio_output.cpp
+++ b/src/tangara/audio/bt_audio_output.cpp
@@ -121,4 +121,8 @@ auto BluetoothAudioOutput::Configure(const Format& fmt) -> void {
   // No configuration necessary; the output format is fixed.
 }
 
+auto BluetoothAudioOutput::samplesUsed() -> uint32_t {
+  return bluetooth_.SamplesUsed();
+}
+
 }  // namespace audio
diff --git a/src/tangara/audio/bt_audio_output.hpp b/src/tangara/audio/bt_audio_output.hpp
index 942715b7..c5681f9a 100644
--- a/src/tangara/audio/bt_audio_output.hpp
+++ b/src/tangara/audio/bt_audio_output.hpp
@@ -45,6 +45,8 @@ class BluetoothAudioOutput : public IAudioOutput {
   auto PrepareFormat(const Format&) -> Format override;
   auto Configure(const Format& format) -> void override;
 
+  auto samplesUsed() -> uint32_t override;
+
   BluetoothAudioOutput(const BluetoothAudioOutput&) = delete;
   BluetoothAudioOutput& operator=(const BluetoothAudioOutput&) = delete;
 
diff --git a/src/tangara/audio/i2s_audio_output.cpp b/src/tangara/audio/i2s_audio_output.cpp
index 20297193..684bfa92 100644
--- a/src/tangara/audio/i2s_audio_output.cpp
+++ b/src/tangara/audio/i2s_audio_output.cpp
@@ -230,4 +230,8 @@ auto I2SAudioOutput::Configure(const Format& fmt) -> void {
   current_config_ = fmt;
 }
 
+auto I2SAudioOutput::samplesUsed() -> uint32_t {
+  return dac_->SamplesUsed();
+}
+
 }  // namespace audio
diff --git a/src/tangara/audio/i2s_audio_output.hpp b/src/tangara/audio/i2s_audio_output.hpp
index f0ce2cf8..bd7d62fb 100644
--- a/src/tangara/audio/i2s_audio_output.hpp
+++ b/src/tangara/audio/i2s_audio_output.hpp
@@ -43,6 +43,8 @@ class I2SAudioOutput : public IAudioOutput {
   auto PrepareFormat(const Format&) -> Format override;
   auto Configure(const Format& format) -> void override;
 
+  auto samplesUsed() -> uint32_t override;
+
   I2SAudioOutput(const I2SAudioOutput&) = delete;
   I2SAudioOutput& operator=(const I2SAudioOutput&) = delete;
 
diff --git a/src/tangara/audio/processor.cpp b/src/tangara/audio/processor.cpp
index 42b678ca..dd96c892 100644
--- a/src/tangara/audio/processor.cpp
+++ b/src/tangara/audio/processor.cpp
@@ -5,10 +5,12 @@
  */
 
 #include "audio/processor.hpp"
+#include <stdint.h>
 
 #include <algorithm>
 #include <cmath>
 #include <cstdint>
+#include <limits>
 
 #include "audio/audio_events.hpp"
 #include "audio/audio_sink.hpp"
@@ -31,14 +33,15 @@ static constexpr std::size_t kSourceBufferLength = kSampleBufferLength * 2;
 
 namespace audio {
 
-SampleProcessor::SampleProcessor()
+SampleProcessor::SampleProcessor(StreamBufferHandle_t sink)
     : commands_(xQueueCreate(1, sizeof(Args))),
       resampler_(nullptr),
       source_(xStreamBufferCreateWithCaps(kSourceBufferLength,
                                           sizeof(sample::Sample) * 2,
                                           MALLOC_CAP_DMA)),
+      sink_(sink),
       leftover_bytes_(0),
-      samples_sunk_(0) {
+      samples_written_(0) {
   input_buffer_ = {
       reinterpret_cast<sample::Sample*>(heap_caps_calloc(
           kSampleBufferLength, sizeof(sample::Sample), MALLOC_CAP_DMA)),
@@ -60,10 +63,12 @@ SampleProcessor::~SampleProcessor() {
 }
 
 auto SampleProcessor::SetOutput(std::shared_ptr<IAudioOutput> output) -> void {
-  // FIXME: We should add synchronisation here, but we should be careful about
-  // not impacting performance given that the output will change only very
-  // rarely (if ever).
-  sink_ = output;
+  assert(xStreamBufferIsEmpty(sink_));
+  // FIXME: We should add synchronisation here, but we should be careful
+  // about not impacting performance given that the output will change only
+  // very rarely (if ever).
+  output_ = output;
+  samples_written_ = output_->samplesUsed();
 }
 
 auto SampleProcessor::beginStream(std::shared_ptr<TrackInfo> track) -> void {
@@ -71,6 +76,7 @@ auto SampleProcessor::beginStream(std::shared_ptr<TrackInfo> track) -> void {
       .track = new std::shared_ptr<TrackInfo>(track),
       .samples_available = 0,
       .is_end_of_stream = false,
+      .clear_buffers = false,
   };
   xQueueSend(commands_, &args, portMAX_DELAY);
 }
@@ -80,16 +86,18 @@ auto SampleProcessor::continueStream(std::span<sample::Sample> input) -> void {
       .track = nullptr,
       .samples_available = input.size(),
       .is_end_of_stream = false,
+      .clear_buffers = false,
   };
   xQueueSend(commands_, &args, portMAX_DELAY);
   xStreamBufferSend(source_, input.data(), input.size_bytes(), portMAX_DELAY);
 }
 
-auto SampleProcessor::endStream() -> void {
+auto SampleProcessor::endStream(bool cancelled) -> void {
   Args args{
       .track = nullptr,
       .samples_available = 0,
       .is_end_of_stream = true,
+      .clear_buffers = cancelled,
   };
   xQueueSend(commands_, &args, portMAX_DELAY);
 }
@@ -108,7 +116,7 @@ auto SampleProcessor::Main() -> void {
       handleContinueStream(args.samples_available);
     }
     if (args.is_end_of_stream) {
-      handleEndStream();
+      handleEndStream(args.clear_buffers);
     }
   }
 }
@@ -116,31 +124,32 @@ auto SampleProcessor::Main() -> void {
 auto SampleProcessor::handleBeginStream(std::shared_ptr<TrackInfo> track)
     -> void {
   if (track->format != source_format_) {
-    resampler_.reset();
     source_format_ = track->format;
+    // The new stream has a different format to the previous stream (or there
+    // was no previous stream).
+    // First, clean up our filters.
+    resampler_.reset();
     leftover_bytes_ = 0;
 
-    auto new_target = sink_->PrepareFormat(track->format);
-    if (new_target != target_format_) {
-      // The new format is different to the old one. Wait for the sink to
-      // drain before continuing.
-      while (!xStreamBufferIsEmpty(sink_->stream())) {
-        ESP_LOGI(kTag, "waiting for sink stream to drain...");
-        // TODO(jacqueline): Get the sink drain ISR to notify us of this
-        // via semaphore instead of busy-ish waiting.
-        vTaskDelay(pdMS_TO_TICKS(10));
-      }
-
-      sink_->Configure(new_target);
+    // If the output is idle, then we can reconfigure it to the closest format
+    // to our new source.
+    // If the output *wasn't* idle, then we can't reconfigure without an
+    // audible gap in playback. So instead, we simply keep the same target
+    // format and begin resampling.
+    if (xStreamBufferIsEmpty(sink_)) {
+      target_format_ = output_->PrepareFormat(track->format);
+      output_->Configure(target_format_);
     }
-    target_format_ = new_target;
   }
 
-  samples_sunk_ = 0;
+  if (xStreamBufferIsEmpty(sink_)) {
+    samples_written_ = output_->samplesUsed();
+  }
+
   events::Audio().Dispatch(internal::StreamStarted{
       .track = track,
-      .src_format = source_format_,
-      .dst_format = target_format_,
+      .sink_format = target_format_,
+      .cue_at_sample = samples_written_,
   });
 }
 
@@ -222,8 +231,8 @@ auto SampleProcessor::handleSamples(std::span<sample::Sample> input) -> size_t {
   return samples_used;
 }
 
-auto SampleProcessor::handleEndStream() -> void {
-  if (resampler_) {
+auto SampleProcessor::handleEndStream(bool clear_bufs) -> void {
+  if (resampler_ && !clear_bufs) {
     size_t read, written;
     std::tie(read, written) = resampler_->Process({}, resampled_buffer_, true);
 
@@ -232,33 +241,31 @@ auto SampleProcessor::handleEndStream() -> void {
     }
   }
 
-  // Send a final update to finish off this stream's samples.
-  if (samples_sunk_ > 0) {
-    events::Audio().Dispatch(internal::StreamUpdate{
-        .samples_sunk = samples_sunk_,
-    });
-    samples_sunk_ = 0;
+  if (clear_bufs) {
+    assert(xStreamBufferReset(sink_));
+    samples_written_ = output_->samplesUsed();
   }
+
+  // FIXME: This discards any leftover samples, but there probably shouldn't be
+  // any leftover samples. Can this be an assert instead?
   leftover_bytes_ = 0;
 
-  events::Audio().Dispatch(internal::StreamEnded{});
+  events::Audio().Dispatch(internal::StreamEnded{
+      .cue_at_sample = samples_written_,
+  });
 }
 
 auto SampleProcessor::sendToSink(std::span<sample::Sample> samples) -> void {
-  // Update the number of samples sunk so far *before* actually sinking them,
-  // since writing to the stream buffer will block when the buffer gets full.
-  samples_sunk_ += samples.size();
-  if (samples_sunk_ >=
-      target_format_.sample_rate * target_format_.num_channels) {
-    events::Audio().Dispatch(internal::StreamUpdate{
-        .samples_sunk = samples_sunk_,
-    });
-    samples_sunk_ = 0;
+  auto data = std::as_bytes(samples);
+  xStreamBufferSend(sink_, data.data(), data.size(), portMAX_DELAY);
+
+  uint32_t samples_before_overflow =
+      std::numeric_limits<uint32_t>::max() - samples_written_;
+  if (samples_before_overflow < samples.size()) {
+    samples_written_ = samples.size() - samples_before_overflow;
+  } else {
+    samples_written_ += samples.size();
   }
-
-  xStreamBufferSend(sink_->stream(),
-                    reinterpret_cast<std::byte*>(samples.data()),
-                    samples.size_bytes(), portMAX_DELAY);
 }
 
 }  // namespace audio
diff --git a/src/tangara/audio/processor.hpp b/src/tangara/audio/processor.hpp
index 8f197d97..1bd6beff 100644
--- a/src/tangara/audio/processor.hpp
+++ b/src/tangara/audio/processor.hpp
@@ -27,21 +27,21 @@ namespace audio {
  */
 class SampleProcessor {
  public:
-  SampleProcessor();
+  SampleProcessor(StreamBufferHandle_t sink);
   ~SampleProcessor();
 
   auto SetOutput(std::shared_ptr<IAudioOutput>) -> void;
 
   auto beginStream(std::shared_ptr<TrackInfo>) -> void;
   auto continueStream(std::span<sample::Sample>) -> void;
-  auto endStream() -> void;
+  auto endStream(bool cancelled) -> void;
 
  private:
   auto Main() -> void;
 
   auto handleBeginStream(std::shared_ptr<TrackInfo>) -> void;
   auto handleContinueStream(size_t samples_available) -> void;
-  auto handleEndStream() -> void;
+  auto handleEndStream(bool cancel) -> void;
 
   auto handleSamples(std::span<sample::Sample>) -> size_t;
 
@@ -51,23 +51,26 @@ class SampleProcessor {
     std::shared_ptr<TrackInfo>* track;
     size_t samples_available;
     bool is_end_of_stream;
+    bool clear_buffers;
   };
   QueueHandle_t commands_;
 
   std::unique_ptr<Resampler> resampler_;
 
   StreamBufferHandle_t source_;
+  StreamBufferHandle_t sink_;
+
   std::span<sample::Sample> input_buffer_;
   std::span<std::byte> input_buffer_as_bytes_;
 
   std::span<sample::Sample> resampled_buffer_;
 
-  std::shared_ptr<IAudioOutput> sink_;
+  std::shared_ptr<IAudioOutput> output_;
   IAudioOutput::Format source_format_;
   IAudioOutput::Format target_format_;
   size_t leftover_bytes_;
 
-  uint32_t samples_sunk_;
+  uint32_t samples_written_;
 };
 
 }  // namespace audio
diff --git a/src/tangara/audio/stream_cues.cpp b/src/tangara/audio/stream_cues.cpp
new file mode 100644
index 00000000..0161fbf4
--- /dev/null
+++ b/src/tangara/audio/stream_cues.cpp
@@ -0,0 +1,57 @@
+/*
+ * Copyright 2024 jacqueline <me@jacqueline.id.au>
+ *
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+#include "audio/stream_cues.hpp"
+
+#include <cstdint>
+#include <memory>
+
+namespace audio {
+
+StreamCues::StreamCues() : now_(0) {}
+
+auto StreamCues::update(uint32_t sample) -> void {
+  if (sample < now_) {
+    // The current time must have overflowed. Deal with any cues between now_
+    // and UINT32_MAX, then proceed as normal.
+    while (!upcoming_.empty() && upcoming_.front().start_at > now_) {
+      current_ = upcoming_.front();
+      upcoming_.pop_front();
+    }
+  }
+  now_ = sample;
+
+  while (!upcoming_.empty() && upcoming_.front().start_at <= now_) {
+    current_ = upcoming_.front();
+    upcoming_.pop_front();
+  }
+}
+
+auto StreamCues::addCue(std::shared_ptr<TrackInfo> track, uint32_t sample)
+    -> void {
+  upcoming_.push_back(Cue{
+      .track = track,
+      .start_at = sample,
+  });
+}
+
+auto StreamCues::current() -> std::pair<std::shared_ptr<TrackInfo>, uint32_t> {
+  if (!current_) {
+    return {};
+  }
+
+  uint32_t duration;
+  if (now_ < current_->start_at) {
+    // now_ overflowed since this track started.
+    duration = now_ + (UINT32_MAX - current_->start_at);
+  } else {
+    duration = now_ - current_->start_at;
+  }
+
+  return {current_->track, duration};
+}
+
+}  // namespace audio
diff --git a/src/tangara/audio/stream_cues.hpp b/src/tangara/audio/stream_cues.hpp
new file mode 100644
index 00000000..f55865b7
--- /dev/null
+++ b/src/tangara/audio/stream_cues.hpp
@@ -0,0 +1,46 @@
+/*
+ * Copyright 2024 jacqueline <me@jacqueline.id.au>
+ *
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+#pragma once
+
+#include <stdint.h>
+#include <cstdint>
+#include <deque>
+#include <memory>
+
+#include "audio/audio_events.hpp"
+
+namespace audio {
+
+/*
+ * Utility for tracking which track is currently being played (and how long it
+ * has been playing for) based on counting samples that are put into and taken
+ * out of the audio processor's output buffer.
+ */
+class StreamCues {
+ public:
+  StreamCues();
+
+  /* Updates the current track given the new most recently played sample. */
+  auto update(uint32_t sample) -> void;
+  /* Returns the current track, and how long it has been playing for. */
+  auto current() -> std::pair<std::shared_ptr<TrackInfo>, uint32_t>;
+
+  auto addCue(std::shared_ptr<TrackInfo>, uint32_t start_at) -> void;
+
+ private:
+  uint32_t now_;
+
+  struct Cue {
+    std::shared_ptr<TrackInfo> track;
+    uint32_t start_at;
+  };
+
+  std::optional<Cue> current_;
+  std::deque<Cue> upcoming_;
+};
+
+}  // namespace audio