diff --git a/src/dusk/audio/audio.c b/src/dusk/audio/audio.c
index 07c20307..f941c442 100644
--- a/src/dusk/audio/audio.c
+++ b/src/dusk/audio/audio.c
@@ -6,6 +6,7 @@
  */
 
 #include "audio.h"
+#include "audio/mixer/audiomixer.h"
 #include "util/memory.h"
 #include "assert/assert.h"
 
@@ -15,6 +16,7 @@ errorret_t audioInit() {
   memoryZero(&AUDIO, sizeof(audio_t));
 
   errorChain(audioPlatformInit());
+  errorChain(audioMixerInit());
 
   errorOk();
 }
@@ -26,7 +28,7 @@ audiostream_t * audioAquireStream(assetentry_t *asset) {
   // too catches the mistake as early as possible, before any stream slot
   // is handed out for it.
   assertTrue(
-    asset->type == ASSET_LOADER_TYPE_WAV || asset->type == ASSET_LOADER_TYPE_MP3,
+    audioStreamTypeForAssetType(asset->type) != AUDIO_STREAM_TYPE_NULL,
     "Unsupported asset type for an audio stream."
   );
 
@@ -40,8 +42,9 @@ audiostream_t * audioAquireStream(assetentry_t *asset) {
   return NULL;
 }
 
-errorret_t audioUpdate() {
+errorret_t audioUpdateEarly() {
   errorChain(audioPlatformUpdate());
+  errorChain(audioMixerUpdateEarly());
 
   for(uint8_t i = 0; i < AUDIO_STREAMS_MAX; i++) {
     audiostream_t *stream = &AUDIO.streams[i];
@@ -51,12 +54,19 @@ errorret_t audioUpdate() {
   errorOk();
 }
 
+errorret_t audioUpdateLate() {
+  errorChain(audioMixerUpdateLate());
+
+  errorOk();
+}
+
 errorret_t audioDispose() {
   for(uint8_t i = 0; i < AUDIO_STREAMS_MAX; i++) {
     audiostream_t *stream = &AUDIO.streams[i];
     errorChain(audioStreamDispose(stream));
   }
 
+  errorChain(audioMixerDispose());
   errorChain(audioPlatformDispose());
 
   errorOk();
diff --git a/src/dusk/audio/audio.h b/src/dusk/audio/audio.h
index 73e23aac..3ded86b5 100644
--- a/src/dusk/audio/audio.h
+++ b/src/dusk/audio/audio.h
@@ -51,11 +51,23 @@ audiostream_t * audioAquireStream(assetentry_t *asset);
 
 /**
  * Updates the audio subsystem, updating every active stream. Should be
- * called once per frame.
+ * called once per frame, before rendering - keeps stream buffering/decode
+ * work off the render thread's critical path (see duskmad's MP3 decode
+ * ring for why this matters).
  *
  * @return Error indicating success or failure of the operation.
  */
-errorret_t audioUpdate();
+errorret_t audioUpdateEarly();
+
+/**
+ * Updates the audio subsystem's end-of-frame work - currently just starting
+ * whatever the audio mixer queued this frame (see
+ * audioMixerUpdateLate()'s own comment). Should be called once per frame,
+ * after rendering.
+ *
+ * @return Error indicating success or failure of the operation.
+ */
+errorret_t audioUpdateLate();
 
 /**
  * Disposes the audio subsystem, stopping and disposing every active stream.
diff --git a/src/dusk/audio/mixer/audiomixer.c b/src/dusk/audio/mixer/audiomixer.c
index 778dba03..5e4aa893 100644
--- a/src/dusk/audio/mixer/audiomixer.c
+++ b/src/dusk/audio/mixer/audiomixer.c
@@ -6,6 +6,7 @@
  */
 
 #include "audiomixer.h"
+#include "audio/stream/audiostream.h"
 #include "util/memory.h"
 #include "util/string.h"
 #include "assert/assert.h"
@@ -18,7 +19,11 @@ errorret_t audioMixerInit() {
   errorOk();
 }
 
-errorret_t audioMixerUpdate() {
+errorret_t audioMixerUpdateEarly() {
+  errorOk();
+}
+
+errorret_t audioMixerUpdateLate() {
   // For each channel.
   audiomixerchannel_t channel = 0;
   do {
@@ -46,8 +51,8 @@ errorret_t audioMixerUpdate() {
 void audioMixerPlay(
   const char_t *file,
   const audiomixerchannel_t channel,
-  const uint8_t volume,
-  const int8_t pan
+  const float_t volume,
+  const float_t pan
 ) {
   audioMixerPlayLooped(file, channel, volume, pan, 0, 0.0f, 0.0f);
 }
@@ -55,16 +60,19 @@ void audioMixerPlay(
 void audioMixerPlayLooped(
   const char_t *file,
   const audiomixerchannel_t channel,
-  const uint8_t volume,
-  const int8_t pan,
+  const float_t volume,
+  const float_t pan,
   const uint8_t loopCount,
   const float_t loopStart,
   const float_t loopTo
 ) {
   assertStrLenMin(file, 1, "File is required, to stop use audioMixerStop");
   assertTrue(channel < AUDIO_MIXER_CHANNEL_COUNT, "Channel is out of range");
-  assertTrue(volume <= 0xFF, "Volume is out of range");
-  assertTrue(volume > 0, "Volume must be greater than 0");
+  assertTrue(volume > 0.0f && volume <= 1.0f, "Volume is out of range");
+  assertTrue(
+    pan >= AUDIO_STREAM_LEFT && pan <= AUDIO_STREAM_RIGHT,
+    "Pan is out of range"
+  );
   assertTrue(loopStart >= 0.0f, "Loop start incorrect");
   assertTrue(loopTo >= 0.0f, "Loop to incorrect");
 
diff --git a/src/dusk/audio/mixer/audiomixer.h b/src/dusk/audio/mixer/audiomixer.h
index 29d560d1..69e10cc9 100644
--- a/src/dusk/audio/mixer/audiomixer.h
+++ b/src/dusk/audio/mixer/audiomixer.h
@@ -25,8 +25,8 @@ typedef struct {
   assetentry_t *asset;
 
   char_t file[AUDIO_PATH_MAX];
-  uint8_t volume;
-  int8_t pan;
+  float_t volume;
+  float_t pan;
   uint8_t loopCount;
   float_t loopStart;
   float_t loopTo;
@@ -45,24 +45,33 @@ extern audiomixer_t AUDIO_MIXER;
 errorret_t audioMixerInit();
 
 /**
- * Updates the audio mixer.
+ * Updates the audio mixer, called before rendering. Reserved for
+ * time-sensitive mixer state (e.g. volume fades) that should be reflected
+ * before this frame's streams are buffered - nothing uses it yet.
  */
-errorret_t audioMixerUpdate();
+errorret_t audioMixerUpdateEarly();
+
+/**
+ * Updates the audio mixer, called at the end of each frame - starts any
+ * sound queued this frame via audioMixerPlay()/audioMixerPlayLooped(), once
+ * every other system has had a chance to queue one.
+ */
+errorret_t audioMixerUpdateLate();
 
 /**
  * Same as audioMixerPlayLooped() but with no looping.
  * 
  * @param file File to be played.
  * @param channel Channel you want to play this sound on.
- * @param volume Volume of the sound, from 0 (silent) to 0xFF (loudest).
+ * @param volume Volume of the sound, from 0.0 (silent) to 1.0 (loudest).
  * @param pan Stereo panning of the sound, from AUDIO_STREAM_LEFT to
  *            AUDIO_STREAM_RIGHT.
  */
 void audioMixerPlay(
   const char_t *file,
   const audiomixerchannel_t channel,
-  const uint8_t volume,
-  const int8_t pan
+  const float_t volume,
+  const float_t pan
 );
 
 /**
@@ -73,12 +82,12 @@ void audioMixerPlay(
  * to queue multiple sounds on the same channel only the last one will be played
  * and will stop any actively playing on that channel currently.
  * 
- * Volume is also mixed based on the channel, so supplying volume 0xFF is then
- * multiplied by the channel's volume, e.g. 0xDD*0xFF = 0xDD.
- * 
+ * Volume is also mixed based on the channel, so supplying volume 1.0 is then
+ * multiplied by the channel's volume, e.g. 0.5*1.0 = 0.5.
+ *
  * @param file File to be played.
  * @param channel Channel you want to play this sound on.
- * @param volume Volume of the sound, from 0 (silent) to 0xFF (loudest).
+ * @param volume Volume of the sound, from 0.0 (silent) to 1.0 (loudest).
  * @param pan Stereo panning of the sound, from AUDIO_STREAM_LEFT to
  *            AUDIO_STREAM_RIGHT.
  * @param loopCount How many times to loop the sound, 0 for infinite looping.
@@ -89,8 +98,8 @@ void audioMixerPlay(
 void audioMixerPlayLooped(
   const char_t *file,
   const audiomixerchannel_t channel,
-  const uint8_t volume,
-  const int8_t pan,
+  const float_t volume,
+  const float_t pan,
   const uint8_t loopCount,
   const float_t loopStart,
   const float_t loopTo
diff --git a/src/dusk/audio/stream/CMakeLists.txt b/src/dusk/audio/stream/CMakeLists.txt
index 00e2a660..81d37fdd 100644
--- a/src/dusk/audio/stream/CMakeLists.txt
+++ b/src/dusk/audio/stream/CMakeLists.txt
@@ -9,4 +9,5 @@ target_sources(${DUSK_LIBRARY_TARGET_NAME}
     audiostream.c
     audiostreampcm.c
     audiostreammp3.c
+    audiostreamtype.c
 )
\ No newline at end of file
diff --git a/src/dusk/audio/stream/audiostream.c b/src/dusk/audio/stream/audiostream.c
index 550faabb..61d720de 100644
--- a/src/dusk/audio/stream/audiostream.c
+++ b/src/dusk/audio/stream/audiostream.c
@@ -16,17 +16,17 @@ errorret_t audioStreamInit(audiostream_t *stream, assetentry_t *asset) {
     asset->state == ASSET_ENTRY_STATE_LOADED,
     "Asset must be loaded before it can back an audio stream."
   );
-  // Only WAV (-> PCM) and MP3 assets can back an audio stream - a caller
-  // passing the wrong kind of asset entry is a programmer error, not
-  // something that can happen from untrusted data, so this is an assert
-  // rather than an errorThrow (see feedback_assert_vs_error convention).
+  // A caller passing an asset type that can't back an audio stream is a
+  // programmer error, not something that can happen from untrusted data, so
+  // this is an assert rather than an errorThrow (see
+  // feedback_assert_vs_error convention).
   assertTrue(
-    asset->type == ASSET_LOADER_TYPE_WAV || asset->type == ASSET_LOADER_TYPE_MP3,
+    audioStreamTypeForAssetType(asset->type) != AUDIO_STREAM_TYPE_NULL,
     "Unsupported asset type for an audio stream."
   );
 
   stream->state = 0;
-  stream->volume = 0xFF;
+  stream->volume = 1.0f;
   stream->directionality = AUDIO_STREAM_CENTER;
   stream->loopStart = -1;
   stream->loopTo = 0;
@@ -38,6 +38,8 @@ errorret_t audioStreamInit(audiostream_t *stream, assetentry_t *asset) {
   stream->onEnd = NULL;
   stream->loopCount = 0;
   stream->lastLoopCount = 0;
+  stream->loopLimit = 0;
+  stream->loopRestartCount = 0;
 
   // Locked for as long as the stream is in use (see audiostream_t.asset's
   // own comment) - released in audioStreamDispose().
@@ -146,10 +148,11 @@ void audioStreamGetPanFactors(
   assertNotNull(outLeft, "outLeft cannot be NULL.");
   assertNotNull(outRight, "outRight cannot be NULL.");
 
-  // directionality is an int8_t clamped to AUDIO_STREAM_LEFT..RIGHT
-  // (-128..127) by its own type, so pan is always within [-1.0, 0.992] -
-  // no further clamping needed.
-  const float_t pan = (float_t) stream->directionality / 128.0f;
+  // directionality is already a pan factor in AUDIO_STREAM_LEFT..RIGHT
+  // (-1.0..1.0) - assertTrue'd here rather than clamped, since a caller
+  // passing something outside that range is a programmer error (see
+  // audioStreamSetDirectionality()'s own assert).
+  const float_t pan = stream->directionality;
 
   *outLeft = pan > 0 ? (1.0f - pan) : 1.0f;
   *outRight = pan < 0 ? (1.0f + pan) : 1.0f;
@@ -173,17 +176,22 @@ size_t audioStreamComputeEndFrame(
   return endFrame;
 }
 
-void audioStreamSetVolume(audiostream_t *stream, const uint8_t volume) {
+void audioStreamSetVolume(audiostream_t *stream, const float_t volume) {
   assertNotNull(stream, "Stream cannot be NULL.");
+  assertTrue(volume >= 0.0f && volume <= 1.0f, "Volume is out of range.");
   stream->volume = volume;
   // TODO: Do I need to update the device output? PSP may require this
 }
 
 void audioStreamSetDirectionality(
   audiostream_t *stream,
-  const int8_t directionality
+  const float_t directionality
 ) {
   assertNotNull(stream, "Stream cannot be NULL.");
+  assertTrue(
+    directionality >= AUDIO_STREAM_LEFT && directionality <= AUDIO_STREAM_RIGHT,
+    "Directionality is out of range."
+  );
   stream->directionality = directionality;
   // TODO: Need to update internal decoder?
 }
@@ -198,6 +206,12 @@ void audioStreamSetLooping(audiostream_t *stream, const bool_t looping) {
   }
 }
 
+void audioStreamSetLoopLimit(audiostream_t *stream, const uint32_t loopLimit) {
+  assertNotNull(stream, "Stream cannot be NULL.");
+  stream->loopLimit = loopLimit;
+  stream->loopRestartCount = 0;
+}
+
 errorret_t audioStreamUpdate(audiostream_t *stream) {
   assertNotNull(stream, "Stream cannot be NULL.");
 
@@ -232,13 +246,22 @@ errorret_t audioStreamUpdate(audiostream_t *stream) {
     (stream->state & AUDIO_STREAM_STATE_BUFFERED) &&
     audioStreamPlatformIsFinished(stream)
   ) {
-    if(stream->state & AUDIO_STREAM_STATE_LOOPING) {
+    // A limit of 0 means unlimited; otherwise this pass's restart only goes
+    // ahead while fewer than loopLimit restarts have already happened -
+    // once it's reached, fall through to the same stop/onEnd path as a
+    // non-looping stream instead of restarting again.
+    const bool_t loopLimitReached = stream->loopLimit > 0 &&
+      stream->loopRestartCount >= stream->loopLimit;
+
+    if((stream->state & AUDIO_STREAM_STATE_LOOPING) && !loopLimitReached) {
       stream->state &= ~AUDIO_STREAM_STATE_BUFFERED;
+      stream->loopRestartCount++;
 
       // Resume from loopTo rather than the very start of the buffer - only
-      // matters for platforms that reach this generic restart path at all;
-      // PSP/Dolphin loop entirely on their own (thread/hardware) and never
-      // report "finished" while looping, so in practice this is Linux-only.
+      // matters for platforms that reach this generic restart path at all
+      // (Linux and Dolphin, both of which re-buffer per pass and report
+      // "finished" at every loop boundary); PSP loops entirely on its own
+      // read-ahead ring instead (see loopRestartCount's own comment).
       stream->startFrame = (size_t) (stream->loopTo * stream->sampleRate);
 
       if(stream->onLoop != NULL) {
diff --git a/src/dusk/audio/stream/audiostream.h b/src/dusk/audio/stream/audiostream.h
index e7eb4371..0886f5fd 100644
--- a/src/dusk/audio/stream/audiostream.h
+++ b/src/dusk/audio/stream/audiostream.h
@@ -6,6 +6,7 @@
  */
 
 #pragma once
+#include "audio/stream/audiostreamtype.h"
 #include "audio/stream/audiostreampcm.h"
 #include "audio/stream/audiostreammp3.h"
 #include "audio/audiostreamplatform.h"
@@ -31,16 +32,9 @@
 #define AUDIO_STREAM_STATE_PLAYING (1 << 0)
 #define AUDIO_STREAM_STATE_LOOPING (1 << 1)
 #define AUDIO_STREAM_STATE_BUFFERED (1 << 2)
-#define AUDIO_STREAM_CENTER 0
-#define AUDIO_STREAM_LEFT -128
-#define AUDIO_STREAM_RIGHT 127
-
-typedef enum {
-  AUDIO_STREAM_TYPE_NULL,
-  AUDIO_STREAM_TYPE_PCM,
-  AUDIO_STREAM_TYPE_MP3,
-  AUDIO_STREAM_TYPE_COUNT
-} audistreamtype_t;
+#define AUDIO_STREAM_CENTER 0.0f
+#define AUDIO_STREAM_LEFT -1.0f
+#define AUDIO_STREAM_RIGHT 1.0f
 
 typedef struct audiostream_s audiostream_t;
 
@@ -60,12 +54,14 @@ typedef struct audiostream_s {
   // playback may also be holding.
   assetentry_t *asset;
 
-  // Loudness. Can only be 0 to 0xFF
-  uint8_t volume;
+  // Loudness, from 0.0 (silent) to 1.0 (loudest).
+  float_t volume;
 
-  // In stereo space, where do we send the audio.
+  // In stereo space, where do we send the audio - from AUDIO_STREAM_LEFT
+  // (-1.0) to AUDIO_STREAM_RIGHT (1.0), AUDIO_STREAM_CENTER (0.0) by
+  // default.
   // TODO: Can we use a 3D vector + Pro Logic II?
-  int8_t directionality;
+  float_t directionality;
 
   // In seconds, where the loop segment ends and playback jumps back to
   // loopTo - -1 (the default) means "at the end of the stream", i.e. loop
@@ -119,6 +115,22 @@ typedef struct audiostream_s {
   // onLoop for - only ever touched from the main thread.
   uint32_t lastLoopCount;
 
+  // Maximum number of times this stream is allowed to loop before stopping
+  // (firing onEnd) instead of restarting again - 0 (the default) means loop
+  // forever. Has no effect unless looping is also enabled via
+  // audioStreamSetLooping(). Set via audioStreamSetLoopLimit().
+  uint32_t loopLimit;
+
+  // How many loop restarts this stream has already committed to - compared
+  // against loopLimit each time a restart is considered, then incremented
+  // if it goes ahead; reset to 0 by audioStreamSetLoopLimit(). Distinct
+  // from loopCount (which tracks when a loop becomes audible, purely for
+  // onLoop notification, and is PSP-only) - this counts decided restarts,
+  // main-thread-only on every platform (PSP decides whether to loop from
+  // its own read-ahead ring, itself driven from the main thread via
+  // audioStreamPSPIsFinished() - see its own comment).
+  uint32_t loopRestartCount;
+
   // PCM format of the stream's decoded output - read identically by every
   // platform backend regardless of stream type (PCM decodes straight
   // through; MP3 decodes to this same format), so it lives here rather
@@ -286,9 +298,9 @@ size_t audioStreamComputeEndFrame(
  * Sets the playback volume of the given audio stream.
  *
  * @param stream The audio stream to update.
- * @param volume The new volume, from 0 (silent) to 0xFF (loudest).
+ * @param volume The new volume, from 0.0 (silent) to 1.0 (loudest).
  */
-void audioStreamSetVolume(audiostream_t *stream, const uint8_t volume);
+void audioStreamSetVolume(audiostream_t *stream, const float_t volume);
 
 /**
  * Sets the stereo directionality (panning) of the given audio stream.
@@ -299,7 +311,7 @@ void audioStreamSetVolume(audiostream_t *stream, const uint8_t volume);
  */
 void audioStreamSetDirectionality(
   audiostream_t *stream,
-  const int8_t directionality
+  const float_t directionality
 );
 
 /**
@@ -313,6 +325,19 @@ void audioStreamSetDirectionality(
  */
 void audioStreamSetLooping(audiostream_t *stream, const bool_t looping);
 
+/**
+ * Sets the maximum number of times the given audio stream is allowed to
+ * loop before stopping (firing onEnd) instead of restarting again. Has no
+ * effect unless looping is also enabled via audioStreamSetLooping(). Resets
+ * the stream's current loop-restart progress back to 0, so the stream gets
+ * the full count from this call onward.
+ *
+ * @param stream The audio stream to update.
+ * @param loopLimit Maximum number of loop restarts, or 0 (the default) to
+ *                   loop forever.
+ */
+void audioStreamSetLoopLimit(audiostream_t *stream, const uint32_t loopLimit);
+
 /**
  * Updates the given audio stream, decoding new data and advancing playback
  * as needed. Should be called every frame for every active stream.
diff --git a/src/dusk/audio/stream/audiostreamtype.c b/src/dusk/audio/stream/audiostreamtype.c
new file mode 100644
index 00000000..5dbc8e2a
--- /dev/null
+++ b/src/dusk/audio/stream/audiostreamtype.c
@@ -0,0 +1,41 @@
+/**
+ * Copyright (c) 2026 Dominic Masters
+ *
+ * This software is released under the MIT License.
+ * https://opensource.org/licenses/MIT
+ */
+
+#include "audiostreamtype.h"
+#include "util/string.h"
+
+const audiostreamassettypeinfo_t AUDIO_STREAM_TYPE_FOR_ASSET_TYPE[ASSET_LOADER_TYPE_COUNT] = {
+  [ASSET_LOADER_TYPE_WAV] = { .type = AUDIO_STREAM_TYPE_PCM, .extension = ".wav" },
+  [ASSET_LOADER_TYPE_MP3] = { .type = AUDIO_STREAM_TYPE_MP3, .extension = ".mp3" }
+};
+
+assetloadertype_t audioStreamAssetTypeForPath(const char_t *path) {
+  for(assetloadertype_t i = 0; i < ASSET_LOADER_TYPE_COUNT; i++) {
+    const char_t *extension = AUDIO_STREAM_TYPE_FOR_ASSET_TYPE[i].extension;
+    if(extension != NULL && stringEndsWithCaseInsensitive(path, extension)) {
+      return i;
+    }
+  }
+
+  return ASSET_LOADER_TYPE_NULL;
+}
+
+audistreamtype_t audioStreamTypeForAssetType(const assetloadertype_t assetType) {
+  return AUDIO_STREAM_TYPE_FOR_ASSET_TYPE[assetType].type;
+}
+
+assetloadertype_t audioStreamAssetTypeForStreamType(const audistreamtype_t streamType) {
+  if(streamType == AUDIO_STREAM_TYPE_NULL) return ASSET_LOADER_TYPE_NULL;
+
+  for(assetloadertype_t i = 0; i < ASSET_LOADER_TYPE_COUNT; i++) {
+    if(AUDIO_STREAM_TYPE_FOR_ASSET_TYPE[i].type == streamType) {
+      return i;
+    }
+  }
+
+  return ASSET_LOADER_TYPE_NULL;
+}
diff --git a/src/dusk/audio/stream/audiostreamtype.h b/src/dusk/audio/stream/audiostreamtype.h
new file mode 100644
index 00000000..14c237a2
--- /dev/null
+++ b/src/dusk/audio/stream/audiostreamtype.h
@@ -0,0 +1,68 @@
+/**
+ * Copyright (c) 2026 Dominic Masters
+ *
+ * This software is released under the MIT License.
+ * https://opensource.org/licenses/MIT
+ */
+
+#pragma once
+#include "dusk.h"
+#include "asset/loader/assetloader.h"
+
+typedef enum {
+  AUDIO_STREAM_TYPE_NULL,
+  AUDIO_STREAM_TYPE_PCM,
+  AUDIO_STREAM_TYPE_MP3,
+  AUDIO_STREAM_TYPE_COUNT
+} audistreamtype_t;
+
+typedef struct {
+  // The audistreamtype_t an audio stream backed by this asset type should
+  // become - AUDIO_STREAM_TYPE_NULL (the zero-initialized default) for any
+  // asset type that can't back an audio stream.
+  audistreamtype_t type;
+
+  // This asset type's file extension, including the leading '.' (e.g.
+  // ".wav"). NULL for any asset type that can't back an audio stream.
+  const char_t *extension;
+} audiostreamassettypeinfo_t;
+
+// Maps an assetloadertype_t to the info describing the audio stream it
+// backs. Indexed directly by assetentry_t.type; see audioStreamInit()'s own
+// use of this for the canonical "is this asset type supported" check. Use
+// the functions below rather than indexing this directly, except where
+// you already have an assetloadertype_t in hand.
+extern const audiostreamassettypeinfo_t AUDIO_STREAM_TYPE_FOR_ASSET_TYPE[ASSET_LOADER_TYPE_COUNT];
+
+/**
+ * Determines which asset loader type a file path's extension corresponds
+ * to, by matching it against AUDIO_STREAM_TYPE_FOR_ASSET_TYPE's own
+ * extensions (case-insensitively).
+ *
+ * @param path The file path to inspect, e.g. "audio/boa.mp3".
+ * @return The matching assetloadertype_t, or ASSET_LOADER_TYPE_NULL if the
+ *         path's extension doesn't match any audio asset type.
+ */
+assetloadertype_t audioStreamAssetTypeForPath(const char_t *path);
+
+/**
+ * Looks up the audistreamtype_t an audio stream backed by the given asset
+ * type should become.
+ *
+ * @param assetType The asset loader type to look up.
+ * @return The matching audistreamtype_t, or AUDIO_STREAM_TYPE_NULL if
+ *         assetType can't back an audio stream.
+ */
+audistreamtype_t audioStreamTypeForAssetType(const assetloadertype_t assetType);
+
+/**
+ * Looks up which asset loader type backs audio streams of the given type -
+ * the reverse of audioStreamTypeForAssetType(). Scans
+ * AUDIO_STREAM_TYPE_FOR_ASSET_TYPE for the first entry whose type matches,
+ * since it isn't itself indexed by audistreamtype_t.
+ *
+ * @param streamType The audio stream type to look up.
+ * @return The matching assetloadertype_t, or ASSET_LOADER_TYPE_NULL if
+ *         streamType is AUDIO_STREAM_TYPE_NULL or otherwise unmapped.
+ */
+assetloadertype_t audioStreamAssetTypeForStreamType(const audistreamtype_t streamType);
diff --git a/src/dusk/engine/engine.c b/src/dusk/engine/engine.c
index 4ed3d426..856c2e51 100644
--- a/src/dusk/engine/engine.c
+++ b/src/dusk/engine/engine.c
@@ -99,7 +99,7 @@ errorret_t engineUpdate(void) {
   timeUpdate();
   inputUpdate();
   consoleUpdate();
-  errorChain(audioUpdate());
+  errorChain(audioUpdateEarly());
   errorChain(rpgUpdate());
   errorChain(sceneUpdate());
   errorChain(assetUpdate());
@@ -107,6 +107,7 @@ errorret_t engineUpdate(void) {
 
   // Render
   errorChain(displayUpdate());
+  errorChain(audioUpdateLate());
   if(inputPressed(INPUT_ACTION_RAGEQUIT)) ENGINE.running = false;
   errorOk();
 }
diff --git a/src/duskdolphin/audio/audiostreamdolphin.c b/src/duskdolphin/audio/audiostreamdolphin.c
index a54ffbab..cfb1847c 100644
--- a/src/duskdolphin/audio/audiostreamdolphin.c
+++ b/src/duskdolphin/audio/audiostreamdolphin.c
@@ -74,7 +74,7 @@ errorret_t audioStreamDolphinBuffer(audiostream_t *stream) {
 
   float_t leftFactor, rightFactor;
   audioStreamGetPanFactors(stream, &leftFactor, &rightFactor);
-  const float_t baseVolume = (float_t) stream->volume / 255.0f;
+  const float_t baseVolume = stream->volume;
   stream->platform.volumeLeft = (s32) mathClamp(baseVolume * leftFactor * MAX_VOLUME, 0, MAX_VOLUME);
   stream->platform.volumeRight = (s32) mathClamp(baseVolume * rightFactor * MAX_VOLUME, 0, MAX_VOLUME);
 
diff --git a/src/dusklinux/audio/audiostreamlinux.c b/src/dusklinux/audio/audiostreamlinux.c
index ac963946..978908ef 100644
--- a/src/dusklinux/audio/audiostreamlinux.c
+++ b/src/dusklinux/audio/audiostreamlinux.c
@@ -191,7 +191,7 @@ errorret_t audioStreamLinuxFeed(audiostream_t *stream) {
   const size_t bytesRead = framesRead * frameSize;
   int queued;
 
-  if(stream->volume == 0xFF) {
+  if(stream->volume >= 1.0f) {
     // Nothing to mix at full volume - queue the window directly instead of
     // allocating/zeroing a same-size scratch buffer just to copy it in.
     queued = SDL_QueueAudio(stream->platform.device, chunk, (Uint32) bytesRead);
@@ -200,7 +200,7 @@ errorret_t audioStreamLinuxFeed(audiostream_t *stream) {
     memoryZero(mixed, bytesRead);
     SDL_MixAudioFormat(
       mixed, (uint8_t *) chunk, AUDIO_S16SYS, (Uint32) bytesRead,
-      (stream->volume * SDL_MIX_MAXVOLUME) / 0xFF
+      (int) (stream->volume * SDL_MIX_MAXVOLUME)
     );
     queued = SDL_QueueAudio(stream->platform.device, mixed, (Uint32) bytesRead);
     memoryFree(mixed);
diff --git a/src/duskpsp/audio/audiostreampsp.c b/src/duskpsp/audio/audiostreampsp.c
index ecc83681..56ff7a12 100644
--- a/src/duskpsp/audio/audiostreampsp.c
+++ b/src/duskpsp/audio/audiostreampsp.c
@@ -232,9 +232,15 @@ void audioStreamPSPTopUp(audiostream_t *stream) {
   // detection ("racing through content"). A short read that isn't
   // genuinely the end just means less got queued this call; the next
   // TopUp() call (next engine frame) naturally retries.
+
+  // A limit of 0 means unlimited - see loopRestartCount's own comment for
+  // why counting decided (not yet necessarily audible) restarts here is
+  // still the right thing to compare against loopLimit.
+  const bool_t loopLimitReached = stream->loopLimit > 0 &&
+    stream->loopRestartCount >= stream->loopLimit;
   const bool_t reachesSegmentEnd = stream->platform.readPosition >= currentEndFrame;
   const bool_t willLoop = reachesSegmentEnd &&
-    (stream->state & AUDIO_STREAM_STATE_LOOPING);
+    (stream->state & AUDIO_STREAM_STATE_LOOPING) && !loopLimitReached;
 
   if(reachesSegmentEnd) {
     if(willLoop) {
@@ -243,6 +249,7 @@ void audioStreamPSPTopUp(audiostream_t *stream) {
         return;
       }
       stream->platform.readPosition = stream->platform.loopToFrame;
+      stream->loopRestartCount++;
 
       threadMutexLock(&stream->platform.ringLock);
       if(stream->platform.loopMarkerCount < AUDIO_PSP_LOOP_MARKER_MAX) {
@@ -379,7 +386,7 @@ void audioStreamPSPThreadFeed(thread_t *thread) {
       float_t leftFactor, rightFactor;
       audioStreamGetPanFactors(stream, &leftFactor, &rightFactor);
 
-      const int baseVolume = (stream->volume * PSP_AUDIO_VOLUME_MAX) / 0xFF;
+      const int baseVolume = (int) (stream->volume * PSP_AUDIO_VOLUME_MAX);
       const int leftVolume = (int) (baseVolume * leftFactor);
       const int rightVolume = (int) (baseVolume * rightFactor);