WIP rewrie audio pipeline+fsm guts for more reliability

16 files changed, 385 insertions, 327 deletions

diff --git a/dependencies.lock b/dependencies.lock
index a9723d1e..f4489997 100644
--- a/dependencies.lock
+++ b/dependencies.lock
@@ -4,6 +4,6 @@ dependencies:
     source:
       type: idf
     version: 5.1.1
-manifest_hash: b9761e0028130d307b778c710e5dd39fb3c942d8084ed429d448d938957fb0e6
+manifest_hash: 9e4320e6f25503854c6c93bcbfa9b80f780485bcf066bdbad31a820544492538
 target: esp32
 version: 1.0.0
diff --git a/src/app_console/app_console.cpp b/src/app_console/app_console.cpp
index 94a48955..7c7c1abc 100644
--- a/src/app_console/app_console.cpp
+++ b/src/app_console/app_console.cpp
@@ -53,10 +53,15 @@ namespace console {
 std::shared_ptr<system_fsm::ServiceLocator> AppConsole::sServices;
 
 int CmdVersion(int argc, char** argv) {
-  std::cout << "firmware-version=" << esp_app_get_description()->version << std::endl;
-  std::cout << "samd-version=" << AppConsole::sServices->samd().Version() << std::endl;
-  std::cout << "collation=" << AppConsole::sServices->collator().Describe().value_or("") << std::endl;
-  std::cout << "database-schema=" << uint32_t(database::kCurrentDbVersion) << std::endl;
+  std::cout << "firmware-version=" << esp_app_get_description()->version
+            << std::endl;
+  std::cout << "samd-version=" << AppConsole::sServices->samd().Version()
+            << std::endl;
+  std::cout << "collation="
+            << AppConsole::sServices->collator().Describe().value_or("")
+            << std::endl;
+  std::cout << "database-schema=" << uint32_t(database::kCurrentDbVersion)
+            << std::endl;
   return 0;
 }
 
@@ -148,7 +153,7 @@ int CmdPlayFile(int argc, char** argv) {
     database::TrackId id = std::atoi(argv[1]);
     AppConsole::sServices->track_queue().append(id);
   } else {
-    std::pmr::string path{&memory::kSpiRamResource};
+    std::string path;
     path += '/';
     path += argv[1];
     for (int i = 2; i < argc; i++) {
@@ -156,8 +161,7 @@ int CmdPlayFile(int argc, char** argv) {
       path += argv[i];
     }
 
-    events::Audio().Dispatch(
-        audio::PlayFile{.filename = {path.data(), path.size()}});
+    events::Audio().Dispatch(audio::SetTrack{.new_track = path});
   }
 
   return 0;
diff --git a/src/audio/audio_converter.cpp b/src/audio/audio_converter.cpp
index 946a0b63..1b233731 100644
--- a/src/audio/audio_converter.cpp
+++ b/src/audio/audio_converter.cpp
@@ -5,14 +5,17 @@
  */
 
 #include "audio_converter.hpp"
+#include <stdint.h>
 
 #include <algorithm>
 #include <cmath>
 #include <cstdint>
 
+#include "audio_events.hpp"
 #include "audio_sink.hpp"
 #include "esp_heap_caps.h"
 #include "esp_log.h"
+#include "event_queue.hpp"
 #include "freertos/portmacro.h"
 #include "freertos/projdefs.h"
 #include "i2s_dac.hpp"
@@ -35,7 +38,9 @@ SampleConverter::SampleConverter()
       resampler_(nullptr),
       source_(xStreamBufferCreateWithCaps(kSourceBufferLength,
                                           sizeof(sample::Sample) * 2,
-                                          MALLOC_CAP_DMA)) {
+                                          MALLOC_CAP_DMA)),
+      leftover_bytes_(0),
+      samples_sunk_(0) {
   input_buffer_ = {
       reinterpret_cast<sample::Sample*>(heap_caps_calloc(
           kSampleBufferLength, sizeof(sample::Sample), MALLOC_CAP_DMA)),
@@ -107,6 +112,19 @@ auto SampleConverter::Main() -> void {
         sink_->Configure(new_target);
       }
       target_format_ = new_target;
+
+      // Send a final sample count for the previous sample rate.
+      if (samples_sunk_ > 0) {
+        events::Audio().Dispatch(internal::ConverterProgress{
+            .samples_sunk = samples_sunk_,
+        });
+      }
+
+      samples_sunk_ = 0;
+      events::Audio().Dispatch(internal::ConverterConfigurationChanged{
+          .src_format = source_format_,
+          .dst_format = target_format_,
+      });
     }
 
     // Loop until we finish reading all the bytes indicated. There might be
@@ -154,9 +172,8 @@ auto SampleConverter::HandleSamples(cpp::span<sample::Sample> input,
   if (source_format_ == target_format_) {
     // The happiest possible case: the input format matches the output
     // format already.
-    std::size_t bytes_sent = xStreamBufferSend(
-        sink_->stream(), input.data(), input.size_bytes(), portMAX_DELAY);
-    return bytes_sent / sizeof(sample::Sample);
+    SendToSink(input);
+    return input.size();
   }
 
   size_t samples_used = 0;
@@ -186,16 +203,26 @@ auto SampleConverter::HandleSamples(cpp::span<sample::Sample> input,
       samples_used = input.size();
     }
 
-    size_t bytes_sent = 0;
-    size_t bytes_to_send = output_source.size_bytes();
-    while (bytes_sent < bytes_to_send) {
-      bytes_sent += xStreamBufferSend(
-          sink_->stream(),
-          reinterpret_cast<std::byte*>(output_source.data()) + bytes_sent,
-          bytes_to_send - bytes_sent, portMAX_DELAY);
-    }
+    SendToSink(output_source);
   }
   return samples_used;
 }
 
+auto SampleConverter::SendToSink(cpp::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::ConverterProgress{
+        .samples_sunk = samples_sunk_,
+    });
+    samples_sunk_ = 0;
+  }
+
+  xStreamBufferSend(sink_->stream(),
+                    reinterpret_cast<std::byte*>(samples.data()),
+                    samples.size_bytes(), portMAX_DELAY);
+}
+
 }  // namespace audio
diff --git a/src/audio/audio_decoder.cpp b/src/audio/audio_decoder.cpp
index 68a8a86b..55ebc0ec 100644
--- a/src/audio/audio_decoder.cpp
+++ b/src/audio/audio_decoder.cpp
@@ -5,6 +5,7 @@
  */
 
 #include "audio_decoder.hpp"
+#include <stdint.h>
 
 #include <cstdint>
 #include <cstdlib>
@@ -50,39 +51,6 @@ namespace audio {
 static constexpr std::size_t kCodecBufferLength =
     drivers::kI2SBufferLengthFrames * sizeof(sample::Sample);
 
-Timer::Timer(std::shared_ptr<Track> t,
-             const codecs::ICodec::OutputFormat& format,
-             uint32_t current_seconds)
-    : track_(t),
-      current_seconds_(current_seconds),
-      current_sample_in_second_(0),
-      samples_per_second_(format.sample_rate_hz * format.num_channels),
-      total_duration_seconds_(format.total_samples.value_or(0) /
-                              format.num_channels / format.sample_rate_hz) {
-  track_->duration = total_duration_seconds_;
-}
-
-auto Timer::AddSamples(std::size_t samples) -> void {
-  bool incremented = false;
-  current_sample_in_second_ += samples;
-  while (current_sample_in_second_ >= samples_per_second_) {
-    current_seconds_++;
-    current_sample_in_second_ -= samples_per_second_;
-    incremented = true;
-  }
-
-  if (incremented) {
-    if (total_duration_seconds_ < current_seconds_) {
-      total_duration_seconds_ = current_seconds_;
-      track_->duration = total_duration_seconds_;
-    }
-
-    PlaybackUpdate ev{.seconds_elapsed = current_seconds_, .track = track_};
-    events::Audio().Dispatch(ev);
-    events::Ui().Dispatch(ev);
-  }
-}
-
 auto Decoder::Start(std::shared_ptr<IAudioSource> source,
                     std::shared_ptr<SampleConverter> sink) -> Decoder* {
   Decoder* task = new Decoder(source, sink);
@@ -92,11 +60,7 @@ auto Decoder::Start(std::shared_ptr<IAudioSource> source,
 
 Decoder::Decoder(std::shared_ptr<IAudioSource> source,
                  std::shared_ptr<SampleConverter> mixer)
-    : source_(source),
-      converter_(mixer),
-      codec_(),
-      timer_(),
-      current_format_() {
+    : source_(source), converter_(mixer), codec_(), current_format_() {
   ESP_LOGI(kTag, "allocating codec buffer, %u KiB", kCodecBufferLength / 1024);
   codec_buffer_ = {
       reinterpret_cast<sample::Sample*>(heap_caps_calloc(
@@ -117,7 +81,6 @@ void Decoder::Main() {
     }
 
     if (ContinueDecoding()) {
-      events::Audio().Dispatch(internal::InputFileFinished{});
       stream_.reset();
     }
   }
@@ -129,6 +92,7 @@ auto Decoder::BeginDecoding(std::shared_ptr<TaggedStream> stream) -> bool {
   codec_.reset(codecs::CreateCodecForType(stream->type()).value_or(nullptr));
   if (!codec_) {
     ESP_LOGE(kTag, "no codec found");
+    events::Audio().Dispatch(internal::DecoderError{});
     return false;
   }
 
@@ -136,6 +100,7 @@ auto Decoder::BeginDecoding(std::shared_ptr<TaggedStream> stream) -> bool {
   if (open_res.has_error()) {
     ESP_LOGE(kTag, "codec failed to start: %s",
              codecs::ICodec::ErrorString(open_res.error()).c_str());
+    events::Audio().Dispatch(internal::DecoderError{});
     return false;
   }
   stream->SetPreambleFinished();
@@ -146,20 +111,23 @@ auto Decoder::BeginDecoding(std::shared_ptr<TaggedStream> stream) -> bool {
   };
 
   ESP_LOGI(kTag, "stream started ok");
-  events::Audio().Dispatch(internal::InputFileOpened{});
-
-  auto tags = std::make_shared<Track>(Track{
-      .tags = stream->tags(),
-      .db_info = {},
-      .bitrate_kbps = open_res->sample_rate_hz,
-      .encoding = stream->type(),
-      .filepath = stream->Filepath(),
-  });
-  timer_.reset(new Timer(tags, open_res.value(), stream->Offset()));
 
-  PlaybackUpdate ev{.seconds_elapsed = stream->Offset(), .track = tags};
-  events::Audio().Dispatch(ev);
-  events::Ui().Dispatch(ev);
+  std::optional<uint32_t> duration;
+  if (open_res->total_samples) {
+    duration = open_res->total_samples.value() / open_res->num_channels /
+               open_res->sample_rate_hz;
+  }
+
+  events::Audio().Dispatch(internal::DecoderOpened{
+      .track = std::make_shared<TrackInfo>(TrackInfo{
+          .tags = stream->tags(),
+          .uri = stream->Filepath(),
+          .duration = duration,
+          .start_offset = stream->Offset(),
+          .bitrate_kbps = open_res->sample_rate_hz,
+          .encoding = stream->type(),
+      }),
+  });
 
   return true;
 }
@@ -167,6 +135,7 @@ auto Decoder::BeginDecoding(std::shared_ptr<TaggedStream> stream) -> bool {
 auto Decoder::ContinueDecoding() -> bool {
   auto res = codec_->DecodeTo(codec_buffer_);
   if (res.has_error()) {
+    events::Audio().Dispatch(internal::DecoderError{});
     return true;
   }
 
@@ -176,11 +145,8 @@ auto Decoder::ContinueDecoding() -> bool {
                                res->is_stream_finished);
   }
 
-  if (timer_) {
-    timer_->AddSamples(res->samples_written);
-  }
-
   if (res->is_stream_finished) {
+    events::Audio().Dispatch(internal::DecoderClosed{});
     codec_.reset();
   }
 
diff --git a/src/audio/audio_fsm.cpp b/src/audio/audio_fsm.cpp
index 05c7c216..7a138cba 100644
--- a/src/audio/audio_fsm.cpp
+++ b/src/audio/audio_fsm.cpp
@@ -36,6 +36,7 @@
 #include "sample.hpp"
 #include "service_locator.hpp"
 #include "system_events.hpp"
+#include "tinyfsm.hpp"
 #include "track.hpp"
 #include "track_queue.hpp"
 #include "wm8523.hpp"
@@ -54,13 +55,158 @@ 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 kDrainBufferSize = sizeof(sample::Sample) * 48000 * 4;
+constexpr size_t kDrainLatencySamples = 48000;
+constexpr size_t kDrainBufferSize =
+    sizeof(sample::Sample) * kDrainLatencySamples * 4;
+
 StreamBufferHandle_t AudioState::sDrainBuffer;
 
-std::optional<database::TrackId> AudioState::sCurrentTrack;
-bool AudioState::sIsPlaybackAllowed;
+std::shared_ptr<TrackInfo> AudioState::sCurrentTrack;
+uint64_t AudioState::sCurrentSamples;
+std::optional<IAudioOutput::Format> AudioState::sCurrentFormat;
+
+std::shared_ptr<TrackInfo> AudioState::sNextTrack;
+uint64_t AudioState::sNextTrackCueSamples;
+
+bool AudioState::sIsResampling;
+bool AudioState::sIsPaused = true;
+
+auto AudioState::currentPositionSeconds() -> std::optional<uint32_t> {
+  if (!sCurrentTrack || !sCurrentFormat) {
+    return {};
+  }
+  return sCurrentSamples /
+         (sCurrentFormat->num_channels * sCurrentFormat->sample_rate);
+}
+
+void AudioState::react(const QueueUpdate& ev) {
+  if (!ev.current_changed && ev.reason != QueueUpdate::kRepeatingLastTrack) {
+    return;
+  }
+
+  SetTrack::Transition transition;
+  switch (ev.reason) {
+    case QueueUpdate::kExplicitUpdate:
+      transition = SetTrack::Transition::kHardCut;
+      break;
+    case QueueUpdate::kRepeatingLastTrack:
+    case QueueUpdate::kTrackFinished:
+      transition = SetTrack::Transition::kGapless;
+      break;
+    case QueueUpdate::kDeserialised:
+    default:
+      // The current track is deserialised separately in order to retain seek
+      // position.
+      return;
+  }
+
+  SetTrack cmd{
+      .new_track = {},
+      .seek_to_second = 0,
+      .transition = transition,
+  };
 
-static std::optional<std::pair<std::string, uint32_t>> sLastTrackUpdate;
+  auto current = sServices->track_queue().current();
+  if (current) {
+    cmd.new_track = *current;
+  }
+
+  tinyfsm::FsmList<AudioState>::dispatch(cmd);
+}
+
+void AudioState::react(const SetTrack& ev) {
+  if (ev.transition == SetTrack::Transition::kHardCut) {
+    clearDrainBuffer();
+  }
+
+  // Move the rest of the work to a background worker, since it may require db
+  // lookups to resolve a track id into a path.
+  auto new_track = ev.new_track;
+  uint32_t seek_to = ev.seek_to_second.value_or(0);
+  sServices->bg_worker().Dispatch<void>([=]() {
+    std::optional<std::string> path;
+    if (std::holds_alternative<database::TrackId>(new_track)) {
+      auto db = sServices->database().lock();
+      if (db) {
+        path = db->getTrackPath(std::get<database::TrackId>(new_track));
+      }
+    } else if (std::holds_alternative<std::string>(new_track)) {
+      path = std::get<std::string>(new_track);
+    }
+
+    if (path) {
+      sFileSource->SetPath(*path, seek_to);
+    } else {
+      sFileSource->SetPath();
+    }
+  });
+}
+
+void AudioState::react(const TogglePlayPause& ev) {
+  sIsPaused = !ev.set_to.value_or(sIsPaused);
+  if (!sIsPaused && is_in_state<states::Standby>() && sCurrentTrack) {
+    transit<states::Playback>();
+  } else if (sIsPaused && is_in_state<states::Playback>()) {
+    transit<states::Standby>();
+  }
+}
+
+void AudioState::react(const internal::DecoderOpened& ev) {
+  ESP_LOGI(kTag, "decoder opened %s", ev.track->uri.c_str());
+  sNextTrack = ev.track;
+  sNextTrackCueSamples = sCurrentSamples + kDrainLatencySamples;
+}
+
+void AudioState::react(const internal::DecoderClosed&) {
+  ESP_LOGI(kTag, "decoder closed");
+  // FIXME: only when we were playing the current track
+  sServices->track_queue().finish();
+}
+
+void AudioState::react(const internal::DecoderError&) {
+  ESP_LOGW(kTag, "decoder errored");
+  // FIXME: only when we were playing the current track
+  sServices->track_queue().finish();
+}
+
+void AudioState::react(const internal::ConverterConfigurationChanged& ev) {
+  sCurrentFormat = ev.dst_format;
+  sIsResampling = ev.src_format != ev.dst_format;
+  ESP_LOGI(kTag, "output format now %u ch @ %lu hz (resample=%i)",
+           sCurrentFormat->num_channels, sCurrentFormat->sample_rate,
+           sIsResampling);
+}
+
+void AudioState::react(const internal::ConverterProgress& ev) {
+  ESP_LOGI(kTag, "sample converter sunk %lu samples", ev.samples_sunk);
+  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) *
+        (sCurrentFormat->num_channels * sCurrentFormat->sample_rate);
+
+    sNextTrack.reset();
+    sNextTrackCueSamples = 0;
+  }
+
+  PlaybackUpdate event{
+      .current_track = sCurrentTrack,
+      .track_position = currentPositionSeconds(),
+      .paused = !is_in_state<states::Playback>(),
+  };
+
+  events::System().Dispatch(event);
+  events::Ui().Dispatch(event);
+
+  if (sCurrentTrack && !sIsPaused && !is_in_state<states::Playback>()) {
+    ESP_LOGI(kTag, "ready to play!");
+    transit<states::Playback>();
+  }
+}
 
 void AudioState::react(const system_fsm::BluetoothEvent& ev) {
   if (ev.event != drivers::bluetooth::Event::kConnectionStateChanged) {
@@ -184,17 +330,6 @@ auto AudioState::clearDrainBuffer() -> void {
   }
 }
 
-auto AudioState::playTrack(database::TrackId id) -> void {
-  sCurrentTrack = id;
-  sServices->bg_worker().Dispatch<void>([=]() {
-    auto db = sServices->database().lock();
-    if (!db) {
-      return;
-    }
-    sFileSource->SetPath(db->getTrackPath(id));
-  });
-}
-
 auto AudioState::commitVolume() -> void {
   auto mode = sServices->nvs().OutputMode();
   auto vol = sOutput->GetVolume();
@@ -209,23 +344,6 @@ auto AudioState::commitVolume() -> void {
   }
 }
 
-auto AudioState::readyToPlay() -> bool {
-  return sCurrentTrack.has_value() && sIsPlaybackAllowed;
-}
-
-void AudioState::react(const TogglePlayPause& ev) {
-  sIsPlaybackAllowed = !sIsPlaybackAllowed;
-  if (readyToPlay()) {
-    if (!is_in_state<states::Playback>()) {
-      transit<states::Playback>();
-    }
-  } else {
-    if (!is_in_state<states::Standby>()) {
-      transit<states::Standby>();
-    }
-  }
-}
-
 namespace states {
 
 void Uninitialised::react(const system_fsm::BootComplete& ev) {
@@ -283,44 +401,6 @@ void Uninitialised::react(const system_fsm::BootComplete& ev) {
   transit<Standby>();
 }
 
-void Standby::react(const PlayFile& ev) {
-  sCurrentTrack = 0;
-  sIsPlaybackAllowed = true;
-  sFileSource->SetPath(ev.filename);
-}
-
-void Playback::react(const PlayFile& ev) {
-  sFileSource->SetPath(ev.filename);
-}
-
-void Standby::react(const SeekFile& ev) {
-  clearDrainBuffer();
-  sFileSource->SetPath(ev.filename, ev.offset);
-}
-
-void Playback::react(const SeekFile& ev) {
-  clearDrainBuffer();
-  sFileSource->SetPath(ev.filename, ev.offset);
-}
-
-void Standby::react(const internal::InputFileOpened& ev) {
-  if (readyToPlay()) {
-    transit<Playback>();
-  }
-}
-
-void Standby::react(const QueueUpdate& ev) {
-  auto current_track = sServices->track_queue().current();
-  if (!current_track || (sCurrentTrack && (*sCurrentTrack == *current_track))) {
-    return;
-  }
-  if (ev.reason == QueueUpdate::Reason::kDeserialised && sLastTrackUpdate) {
-    return;
-  }
-  clearDrainBuffer();
-  playTrack(*current_track);
-}
-
 static const char kQueueKey[] = "audio:queue";
 static const char kCurrentFileKey[] = "audio:current";
 
@@ -328,7 +408,7 @@ void Standby::react(const system_fsm::KeyLockChanged& ev) {
   if (!ev.locking) {
     return;
   }
-  sServices->bg_worker().Dispatch<void>([]() {
+  sServices->bg_worker().Dispatch<void>([this]() {
     auto db = sServices->database().lock();
     if (!db) {
       return;
@@ -341,10 +421,10 @@ void Standby::react(const system_fsm::KeyLockChanged& ev) {
     }
     db->put(kQueueKey, queue.serialise());
 
-    if (sLastTrackUpdate) {
+    if (sCurrentTrack) {
       cppbor::Array current_track{
-          cppbor::Tstr{sLastTrackUpdate->first},
-          cppbor::Uint{sLastTrackUpdate->second},
+          cppbor::Tstr{sCurrentTrack->uri},
+          cppbor::Uint{currentPositionSeconds().value_or(0)},
       };
       db->put(kCurrentFileKey, current_track.toString());
     }
@@ -371,8 +451,12 @@ void Standby::react(const system_fsm::StorageMounted& ev) {
       if (parsed->type() == cppbor::ARRAY) {
         std::string filename = parsed->asArray()->get(0)->asTstr()->value();
         uint32_t pos = parsed->asArray()->get(1)->asUint()->value();
-        sLastTrackUpdate = std::make_pair(filename, pos);
-        sFileSource->SetPath(filename, pos);
+
+        events::Audio().Dispatch(SetTrack{
+            .new_track = filename,
+            .seek_to_second = pos,
+            .transition = SetTrack::Transition::kHardCut,
+        });
       }
     }
 
@@ -388,76 +472,31 @@ void Standby::react(const system_fsm::StorageMounted& ev) {
 }
 
 void Playback::entry() {
-  ESP_LOGI(kTag, "beginning playback");
+  ESP_LOGI(kTag, "audio output resumed");
   sOutput->mode(IAudioOutput::Modes::kOnPlaying);
 
-  events::System().Dispatch(PlaybackStarted{});
-  events::Ui().Dispatch(PlaybackStarted{});
+  PlaybackUpdate event{
+      .current_track = sCurrentTrack,
+      .track_position = currentPositionSeconds(),
+      .paused = false,
+  };
+
+  events::System().Dispatch(event);
+  events::Ui().Dispatch(event);
 }
 
 void Playback::exit() {
-  ESP_LOGI(kTag, "finishing playback");
+  ESP_LOGI(kTag, "audio output paused");
   sOutput->mode(IAudioOutput::Modes::kOnPaused);
 
-  // Stash the current volume now, in case it changed during playback, since
-  // we might be powering off soon.
-  commitVolume();
-
-  events::System().Dispatch(PlaybackStopped{});
-  events::Ui().Dispatch(PlaybackStopped{});
-}
-
-void Playback::react(const system_fsm::HasPhonesChanged& ev) {
-  if (!ev.has_headphones) {
-    transit<Standby>();
-  }
-}
-
-void Playback::react(const QueueUpdate& ev) {
-  if (!ev.current_changed) {
-    return;
-  }
-  // Cut the current track immediately.
-  if (ev.reason == QueueUpdate::Reason::kExplicitUpdate) {
-    clearDrainBuffer();
-  }
-  auto current_track = sServices->track_queue().current();
-  if (!current_track) {
-    sFileSource->SetPath();
-    sCurrentTrack.reset();
-    transit<Standby>();
-    return;
-  }
-  playTrack(*current_track);
-}
-
-void Playback::react(const PlaybackUpdate& ev) {
-  ESP_LOGI(kTag, "elapsed: %lu, total: %lu", ev.seconds_elapsed,
-           ev.track->duration);
-  sLastTrackUpdate = std::make_pair(ev.track->filepath, ev.seconds_elapsed);
-}
-
-void Playback::react(const internal::InputFileOpened& ev) {}
-
-void Playback::react(const internal::InputFileClosed& ev) {}
+  PlaybackUpdate event{
+      .current_track = sCurrentTrack,
+      .track_position = currentPositionSeconds(),
+      .paused = true,
+  };
 
-void Playback::react(const internal::InputFileFinished& ev) {
-  ESP_LOGI(kTag, "finished playing file");
-  sLastTrackUpdate.reset();
-  sServices->track_queue().finish();
-  if (!sServices->track_queue().current()) {
-    for (int i = 0; i < 20; i++) {
-      if (xStreamBufferIsEmpty(sDrainBuffer)) {
-        break;
-      }
-      vTaskDelay(pdMS_TO_TICKS(200));
-    }
-    transit<Standby>();
-  }
-}
-
-void Playback::react(const internal::AudioPipelineIdle& ev) {
-  transit<Standby>();
+  events::System().Dispatch(event);
+  events::Ui().Dispatch(event);
 }
 
 }  // namespace states
diff --git a/src/audio/include/audio_converter.hpp b/src/audio/include/audio_converter.hpp
index c2ebde60..dcd068b5 100644
--- a/src/audio/include/audio_converter.hpp
+++ b/src/audio/include/audio_converter.hpp
@@ -6,6 +6,7 @@
 
 #pragma once
 
+#include <stdint.h>
 #include <cstdint>
 #include <memory>
 
@@ -40,6 +41,8 @@ class SampleConverter {
   auto SetTargetFormat(const IAudioOutput::Format& format) -> void;
   auto HandleSamples(cpp::span<sample::Sample>, bool) -> size_t;
 
+  auto SendToSink(cpp::span<sample::Sample>) -> void;
+
   struct Args {
     IAudioOutput::Format format;
     size_t samples_available;
@@ -59,6 +62,8 @@ class SampleConverter {
   IAudioOutput::Format source_format_;
   IAudioOutput::Format target_format_;
   size_t leftover_bytes_;
+
+  uint32_t samples_sunk_;
 };
 
 }  // namespace audio
diff --git a/src/audio/include/audio_decoder.hpp b/src/audio/include/audio_decoder.hpp
index b8aac710..89f0f43c 100644
--- a/src/audio/include/audio_decoder.hpp
+++ b/src/audio/include/audio_decoder.hpp
@@ -20,25 +20,6 @@
 namespace audio {
 
 /*
- * Sample-based timer for the current elapsed playback time.
- */
-class Timer {
- public:
-  Timer(std::shared_ptr<Track>, const codecs::ICodec::OutputFormat& format, uint32_t current_seconds = 0);
-
-  auto AddSamples(std::size_t) -> void;
-
- private:
-  std::shared_ptr<Track> track_;
-
-  uint32_t current_seconds_;
-  uint32_t current_sample_in_second_;
-  uint32_t samples_per_second_;
-
-  uint32_t total_duration_seconds_;
-};
-
-/*
  * Handle to a persistent task that takes bytes from the given source, decodes
  * them into sample::Sample (normalised to 16 bit signed PCM), and then
  * forwards the resulting stream to the given converter.
@@ -65,7 +46,6 @@ class Decoder {
 
   std::shared_ptr<codecs::IStream> stream_;
   std::unique_ptr<codecs::ICodec> codec_;
-  std::unique_ptr<Timer> timer_;
 
   std::optional<codecs::ICodec::OutputFormat> current_format_;
   std::optional<IAudioOutput::Format> current_sink_format_;
diff --git a/src/audio/include/audio_events.hpp b/src/audio/include/audio_events.hpp
index a8533646..9af30467 100644
--- a/src/audio/include/audio_events.hpp
+++ b/src/audio/include/audio_events.hpp
@@ -9,8 +9,10 @@
 #include <stdint.h>
 #include <cstdint>
 #include <memory>
+#include <optional>
 #include <string>
 
+#include "audio_sink.hpp"
 #include "tinyfsm.hpp"
 
 #include "track.hpp"
@@ -18,24 +20,80 @@
 
 namespace audio {
 
-struct Track {
+/*
+ * Struct encapsulating information about the decoder's current track.
+ */
+struct TrackInfo {
+  /*
+   * Audio tags extracted from the file. May be absent for files without any
+   * parseable tags.
+   */
   std::shared_ptr<database::TrackTags> tags;
-  std::shared_ptr<database::TrackData> db_info;
 
-  uint32_t duration;
-  uint32_t bitrate_kbps;
+  /*
+   * URI that the current track was retrieved from. This is currently always a
+   * file path on the SD card.
+   */
+  std::string uri;
+
+  /*
+   * The length of this track in seconds. This is either retrieved from the
+   * track's tags, or sometimes computed. It may therefore sometimes be
+   * inaccurate or missing.
+   */
+  std::optional<uint32_t> duration;
+
+  /* The offset in seconds that this file's decoding started from. */
+  std::optional<uint32_t> start_offset;
+
+  /* The approximate bitrate of this track in its original encoded form. */
+  std::optional<uint32_t> bitrate_kbps;
+
+  /* The encoded format of the this track. */
   codecs::StreamType encoding;
-  std::string filepath;
 };
 
-struct PlaybackStarted : tinyfsm::Event {};
-
+/*
+ * Event emitted by the audio FSM when the state of the audio pipeline has
+ * changed. This is usually once per second while a track is playing, plus one
+ * event each when a track starts or finishes.
+ */
 struct PlaybackUpdate : tinyfsm::Event {
-  uint32_t seconds_elapsed;
-  std::shared_ptr<Track> track;
+  /*
+   * The track that is currently being decoded by the audio pipeline. May be
+   * absent if there is no current track.
+   */
+  std::shared_ptr<TrackInfo> current_track;
+
+  /*
+   * How long the current track has been playing for, in seconds. Will always
+   * be present if current_track is present.
+   */
+  std::optional<uint32_t> track_position;
+
+  /* Whether or not the current track is currently being output to a sink. */
+  bool paused;
+};
+
+/*
+ * Sets a new track to be decoded by the audio pipeline, replacing any
+ * currently playing track.
+ */
+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 PlaybackStopped : tinyfsm::Event {};
+struct TogglePlayPause : tinyfsm::Event {
+  std::optional<bool> set_to;
+};
 
 struct QueueUpdate : tinyfsm::Event {
   bool current_changed;
@@ -49,15 +107,6 @@ struct QueueUpdate : tinyfsm::Event {
   Reason reason;
 };
 
-struct PlayFile : tinyfsm::Event {
-  std::string filename;
-};
-
-struct SeekFile : tinyfsm::Event {
-  uint32_t offset;
-  std::string filename;
-};
-
 struct StepUpVolume : tinyfsm::Event {};
 struct StepDownVolume : tinyfsm::Event {};
 struct SetVolume : tinyfsm::Event {
@@ -83,17 +132,26 @@ struct SetVolumeLimit : tinyfsm::Event {
   int limit_db;
 };
 
-struct TogglePlayPause : tinyfsm::Event {};
-
 struct OutputModeChanged : tinyfsm::Event {};
 
 namespace internal {
 
-struct InputFileOpened : tinyfsm::Event {};
-struct InputFileClosed : tinyfsm::Event {};
-struct InputFileFinished : tinyfsm::Event {};
+struct DecoderOpened : tinyfsm::Event {
+  std::shared_ptr<TrackInfo> track;
+};
+
+struct DecoderClosed : tinyfsm::Event {};
+
+struct DecoderError : tinyfsm::Event {};
 
-struct AudioPipelineIdle : tinyfsm::Event {};
+struct ConverterConfigurationChanged : tinyfsm::Event {
+  IAudioOutput::Format src_format;
+  IAudioOutput::Format dst_format;
+};
+
+struct ConverterProgress : tinyfsm::Event {
+  uint32_t samples_sunk;
+};
 
 }  // namespace internal
 
diff --git a/src/audio/include/audio_fsm.hpp b/src/audio/include/audio_fsm.hpp
index 13e241be..62bb4786 100644
--- a/src/audio/include/audio_fsm.hpp
+++ b/src/audio/include/audio_fsm.hpp
@@ -6,6 +6,7 @@
 
 #pragma once
 
+#include <stdint.h>
 #include <deque>
 #include <memory>
 #include <vector>
@@ -41,6 +42,17 @@ class AudioState : public tinyfsm::Fsm<AudioState> {
   /* Fallback event handler. Does nothing. */
   void react(const tinyfsm::Event& ev) {}
 
+  void react(const QueueUpdate&);
+  void react(const SetTrack&);
+  void react(const TogglePlayPause&);
+
+  void react(const internal::DecoderOpened&);
+  void react(const internal::DecoderClosed&);
+  void react(const internal::DecoderError&);
+
+  void react(const internal::ConverterConfigurationChanged&);
+  void react(const internal::ConverterProgress&);
+
   void react(const StepUpVolume&);
   void react(const StepDownVolume&);
   virtual void react(const system_fsm::HasPhonesChanged&);
@@ -56,17 +68,6 @@ class AudioState : public tinyfsm::Fsm<AudioState> {
   virtual void react(const system_fsm::StorageMounted&) {}
   virtual void react(const system_fsm::BluetoothEvent&);
 
-  virtual void react(const PlayFile&) {}
-  virtual void react(const SeekFile&) {}
-  virtual void react(const QueueUpdate&) {}
-  virtual void react(const PlaybackUpdate&) {}
-  void react(const TogglePlayPause&);
-
-  virtual void react(const internal::InputFileOpened&) {}
-  virtual void react(const internal::InputFileClosed&) {}
-  virtual void react(const internal::InputFileFinished&) {}
-  virtual void react(const internal::AudioPipelineIdle&) {}
-
  protected:
   auto clearDrainBuffer() -> void;
   auto playTrack(database::TrackId id) -> void;
@@ -83,10 +84,17 @@ class AudioState : public tinyfsm::Fsm<AudioState> {
 
   static StreamBufferHandle_t sDrainBuffer;
 
-  static std::optional<database::TrackId> sCurrentTrack;
+  static std::shared_ptr<TrackInfo> sCurrentTrack;
+  static uint64_t sCurrentSamples;
+  static std::optional<IAudioOutput::Format> sCurrentFormat;
 
-  auto readyToPlay() -> bool;
-  static bool sIsPlaybackAllowed;
+  static std::shared_ptr<TrackInfo> sNextTrack;
+  static uint64_t sNextTrackCueSamples;
+
+  static bool sIsResampling;
+  static bool sIsPaused;
+
+  auto currentPositionSeconds() -> std::optional<uint32_t>;
 };
 
 namespace states {
@@ -94,7 +102,6 @@ namespace states {
 class Uninitialised : public AudioState {
  public:
   void react(const system_fsm::BootComplete&) override;
-
   void react(const system_fsm::BluetoothEvent&) override{};
 
   using AudioState::react;
@@ -102,10 +109,6 @@ class Uninitialised : public AudioState {
 
 class Standby : public AudioState {
  public:
-  void react(const PlayFile&) override;
-  void react(const SeekFile&) override;
-  void react(const internal::InputFileOpened&) override;
-  void react(const QueueUpdate&) override;
   void react(const system_fsm::KeyLockChanged&) override;
   void react(const system_fsm::StorageMounted&) override;
 
@@ -117,18 +120,6 @@ class Playback : public AudioState {
   void entry() override;
   void exit() override;
 
-  void react(const system_fsm::HasPhonesChanged&) override;
-
-  void react(const PlayFile&) override;
-  void react(const SeekFile&) override;
-  void react(const QueueUpdate&) override;
-  void react(const PlaybackUpdate&) override;
-
-  void react(const internal::InputFileOpened&) override;
-  void react(const internal::InputFileClosed&) override;
-  void react(const internal::InputFileFinished&) override;
-  void react(const internal::AudioPipelineIdle&) override;
-
   using AudioState::react;
 };
 
diff --git a/src/audio/track_queue.cpp b/src/audio/track_queue.cpp
index a3f4c815..dbe283c4 100644
--- a/src/audio/track_queue.cpp
+++ b/src/audio/track_queue.cpp
@@ -136,7 +136,7 @@ auto TrackQueue::insert(Item i, size_t index) -> void {
   {
     const std::shared_lock<std::shared_mutex> lock(mutex_);
     was_queue_empty = pos_ == tracks_.size();
-    current_changed = pos_ == was_queue_empty || index == pos_;
+    current_changed = was_queue_empty || index == pos_;
   }
 
   auto update_shuffler = [=, this]() {
diff --git a/src/lua/include/property.hpp b/src/lua/include/property.hpp
index 7d160fba..f19fdeec 100644
--- a/src/lua/include/property.hpp
+++ b/src/lua/include/property.hpp
@@ -23,7 +23,7 @@ using LuaValue = std::variant<std::monostate,
                               int,
                               bool,
                               std::string,
-                              audio::Track,
+                              audio::TrackInfo,
                               drivers::bluetooth::Device,
                               std::vector<drivers::bluetooth::Device>>;
 
diff --git a/src/lua/property.cpp b/src/lua/property.cpp
index f721f9ce..200f4d5c 100644
--- a/src/lua/property.cpp
+++ b/src/lua/property.cpp
@@ -221,7 +221,7 @@ static auto pushTagValue(lua_State* L, const database::TagValue& val) -> void {
       val);
 }
 
-static void pushTrack(lua_State* L, const audio::Track& track) {
+static void pushTrack(lua_State* L, const audio::TrackInfo& track) {
   lua_newtable(L);
 
   for (const auto& tag : track.tags->allPresent()) {
@@ -229,19 +229,18 @@ static void pushTrack(lua_State* L, const audio::Track& track) {
     pushTagValue(L, track.tags->get(tag));
     lua_settable(L, -3);
   }
-  if (track.db_info) {
-    lua_pushliteral(L, "id");
-    lua_pushinteger(L, track.db_info->id);
+
+  if (track.duration) {
+    lua_pushliteral(L, "duration");
+    lua_pushinteger(L, track.duration.value());
     lua_settable(L, -3);
   }
 
-  lua_pushliteral(L, "duration");
-  lua_pushinteger(L, track.duration);
-  lua_settable(L, -3);
-
-  lua_pushliteral(L, "bitrate_kbps");
-  lua_pushinteger(L, track.bitrate_kbps);
-  lua_settable(L, -3);
+  if (track.bitrate_kbps) {
+    lua_pushliteral(L, "bitrate_kbps");
+    lua_pushinteger(L, track.bitrate_kbps.value());
+    lua_settable(L, -3);
+  }
 
   lua_pushliteral(L, "encoding");
   lua_pushstring(L, codecs::StreamTypeToString(track.encoding).c_str());
@@ -289,7 +288,7 @@ auto Property::PushValue(lua_State& s) -> int {
           lua_pushboolean(&s, arg);
         } else if constexpr (std::is_same_v<T, std::string>) {
           lua_pushstring(&s, arg.c_str());
-        } else if constexpr (std::is_same_v<T, audio::Track>) {
+        } else if constexpr (std::is_same_v<T, audio::TrackInfo>) {
           pushTrack(&s, arg);
         } else if constexpr (std::is_same_v<T, drivers::bluetooth::Device>) {
           pushDevice(&s, arg);
diff --git a/src/system_fsm/include/system_fsm.hpp b/src/system_fsm/include/system_fsm.hpp
index cc60e43b..a129829e 100644
--- a/src/system_fsm/include/system_fsm.hpp
+++ b/src/system_fsm/include/system_fsm.hpp
@@ -63,7 +63,7 @@ class SystemState : public tinyfsm::Fsm<SystemState> {
   virtual void react(const SdDetectChanged&) {}
   virtual void react(const SamdUsbMscChanged&) {}
   virtual void react(const database::event::UpdateFinished&) {}
-  virtual void react(const audio::PlaybackStopped&) {}
+  virtual void react(const audio::PlaybackUpdate&) {}
   virtual void react(const internal::IdleTimeout&) {}
   virtual void react(const internal::UnmountTimeout&) {}
 
@@ -101,7 +101,7 @@ class Running : public SystemState {
 
   void react(const KeyLockChanged&) override;
   void react(const SdDetectChanged&) override;
-  void react(const audio::PlaybackStopped&) override;
+  void react(const audio::PlaybackUpdate&) override;
   void react(const database::event::UpdateFinished&) override;
   void react(const SamdUsbMscChanged&) override;
   void react(const internal::UnmountTimeout&) override;
diff --git a/src/system_fsm/running.cpp b/src/system_fsm/running.cpp
index d80809e6..a6ab5d47 100644
--- a/src/system_fsm/running.cpp
+++ b/src/system_fsm/running.cpp
@@ -56,7 +56,7 @@ void Running::react(const KeyLockChanged& ev) {
   checkIdle();
 }
 
-void Running::react(const audio::PlaybackStopped& ev) {
+void Running::react(const audio::PlaybackUpdate& ev) {
   checkIdle();
 }
 
diff --git a/src/ui/include/ui_fsm.hpp b/src/ui/include/ui_fsm.hpp
index f7fde1dd..5e1cc487 100644
--- a/src/ui/include/ui_fsm.hpp
+++ b/src/ui/include/ui_fsm.hpp
@@ -57,8 +57,6 @@ class UiState : public tinyfsm::Fsm<UiState> {
   virtual void react(const system_fsm::StorageMounted&) {}
 
   void react(const system_fsm::BatteryStateChanged&);
-  void react(const audio::PlaybackStarted&);
-  void react(const audio::PlaybackStopped&);
   void react(const audio::PlaybackUpdate&);
   void react(const audio::QueueUpdate&);
 
diff --git a/src/ui/ui_fsm.cpp b/src/ui/ui_fsm.cpp
index a913a339..42c6a99c 100644
--- a/src/ui/ui_fsm.cpp
+++ b/src/ui/ui_fsm.cpp
@@ -114,14 +114,11 @@ lua::Property UiState::sBluetoothDevices{
 
 lua::Property UiState::sPlaybackPlaying{
     false, [](const lua::LuaValue& val) {
-      bool current_val = std::get<bool>(sPlaybackPlaying.Get());
       if (!std::holds_alternative<bool>(val)) {
         return false;
       }
       bool new_val = std::get<bool>(val);
-      if (current_val != new_val) {
-        events::Audio().Dispatch(audio::TogglePlayPause{});
-      }
+      events::Audio().Dispatch(audio::TogglePlayPause{.set_to = new_val});
       return true;
     }};
 
@@ -135,12 +132,13 @@ lua::Property UiState::sPlaybackPosition{
       int new_val = std::get<int>(val);
       if (current_val != new_val) {
         auto track = sPlaybackTrack.Get();
-        if (!std::holds_alternative<audio::Track>(track)) {
+        if (!std::holds_alternative<audio::TrackInfo>(track)) {
           return false;
         }
-        events::Audio().Dispatch(audio::SeekFile{
-            .offset = (uint32_t)new_val,
-            .filename = std::get<audio::Track>(track).filepath});
+        events::Audio().Dispatch(audio::SetTrack{
+            .new_track = std::get<audio::TrackInfo>(track).uri,
+            .seek_to_second = (uint32_t)new_val,
+        });
       }
       return true;
     }};
@@ -393,17 +391,10 @@ void UiState::react(const audio::QueueUpdate&) {
   sQueueReplay.Update(queue.replay());
 }
 
-void UiState::react(const audio::PlaybackStarted& ev) {
-  sPlaybackPlaying.Update(true);
-}
-
 void UiState::react(const audio::PlaybackUpdate& ev) {
-  sPlaybackTrack.Update(*ev.track);
-  sPlaybackPosition.Update(static_cast<int>(ev.seconds_elapsed));
-}
-
-void UiState::react(const audio::PlaybackStopped&) {
-  sPlaybackPlaying.Update(false);
+  sPlaybackTrack.Update(*ev.current_track);
+  sPlaybackPlaying.Update(!ev.paused);
+  sPlaybackPosition.Update(static_cast<int>(ev.track_position.value_or(0)));
 }
 
 void UiState::react(const audio::VolumeChanged& ev) {