diff --git a/include/mixer.h b/include/mixer.h
index 68b29fc..3889e77 100644
--- a/include/mixer.h
+++ b/include/mixer.h
@@ -28,10 +28,25 @@ struct bgi {
 	int volume;
 };
 
+struct wai {
+	//int uk1;
+	//int uk2;
+	int channel;
+	//int uk3; // version 4+
+};
+
 void bgi_read(const char *path);
 struct bgi *bgi_get(int no);
+void wai_load(const char *path);
+struct wai *wai_get(int no);
 
 void mixer_init(void);
+int mixer_get_numof(void);
+const char *mixer_get_name(int n);
+int mixer_get_volume(int n, int *volume);
+int mixer_set_volume(int n, int volume);
+int mixer_get_mute(int n, int *mute);
+int mixer_set_mute(int n, int mute);
 
 struct channel;
 enum asset_type;
diff --git a/include/sts_mixer.h b/include/sts_mixer.h
index 7336eed..8e27d64 100644
--- a/include/sts_mixer.h
+++ b/include/sts_mixer.h
@@ -341,8 +341,9 @@ void sts_mixer_mix_audio(sts_mixer_t* mixer, void* output, unsigned int samples)
     }
 
     // write to buffer
-    left = sts_mixer__clamp_sample(left);
-    right = sts_mixer__clamp_sample(right);
+    // NOTE: xsystem4 change: use mixer gain (not sure why this isn't implemented upstream...)
+    left = sts_mixer__clamp_sample(left * mixer->gain);
+    right = sts_mixer__clamp_sample(right * mixer->gain);
     switch (mixer->audio_format) {
       case STS_MIXER_SAMPLE_FORMAT_8:
         *out_8++ = (char)(left * 127.0f);
diff --git a/include/xsystem4.h b/include/xsystem4.h
index eb66711..a626379 100644
--- a/include/xsystem4.h
+++ b/include/xsystem4.h
@@ -30,6 +30,8 @@ struct config {
 	int view_height;
 	size_t mixer_nr_channels;
 	char **mixer_channels;
+	int *mixer_volumes;
+	int default_volume;
 
 	char *bgi_path;
 	char *wai_path;
diff --git a/src/audio_meta.c b/src/audio_meta.c
index 0b9e32f..0dc316c 100644
--- a/src/audio_meta.c
+++ b/src/audio_meta.c
@@ -15,8 +15,11 @@
  */
 
 #include <stdio.h>
+#include <stdlib.h>
 #include "system4.h"
+#include "system4/file.h"
 #include "mixer.h"
+#include "little_endian.h"
 
 #define BGI_MAX 100
 
@@ -91,3 +94,59 @@ struct bgi *bgi_get(int no)
 	}
 	return NULL;
 }
+
+static struct wai *wai_data = NULL;
+static int nr_wai = 0;
+
+void wai_load(const char *path)
+{
+	size_t len;
+	uint8_t *file = file_read(path, &len);
+
+	if (len < 48 || file[0] != 'X' || file[1] != 'I' || file[2] != '2' || file[3] != '\0') {
+		WARNING("Not a .wai file: %s", path);
+		goto end;
+	}
+	int count = LittleEndian_getDW(file, 8) - 1;
+	if (count < 0) {
+		WARNING("Invalid .wai count: %d", count);
+		goto end;
+	}
+	int version = LittleEndian_getDW(file, 12);
+	if (version < 3 || version > 4) {
+		WARNING("Unsupported .wai version: %d", version);
+		goto end;
+	}
+
+	uint8_t *data;
+	int record_size;
+	if (version == 4) {
+		data = file + 40;
+		record_size = 4*4;
+	} else {
+		data = file + 36;
+		record_size = 3*4;
+	}
+
+	if (len - (version == 4 ? 40 : 36) < (size_t)count*record_size) {
+		WARNING(".wai file truncated");
+		goto end;
+	}
+
+	wai_data = xcalloc(count, sizeof(struct wai));
+	for (int i = 0; i < count; i++) {
+		wai_data[i].channel = LittleEndian_getDW(data, i*record_size + 8);
+	}
+	nr_wai = count;
+
+end:
+	free(file);
+}
+
+struct wai *wai_get(int no)
+{
+	no--;
+	if (no < 0 || no >= nr_wai)
+		return NULL;
+	return &wai_data[no];
+}
diff --git a/src/audio_mixer.c b/src/audio_mixer.c
index 8a6cd16..8e5126e 100644
--- a/src/audio_mixer.c
+++ b/src/audio_mixer.c
@@ -28,6 +28,8 @@
 #include "mixer.h"
 #include "xsystem4.h"
 
+#define clamp(min_value, max_value, value) min(max_value, max(min_value, value))
+
 /*
  * The actual mixer implementation is contained in this header.
  * The rest of this file implements a high level interface for managing mixer
@@ -51,6 +53,7 @@ struct channel {
 	// archive data
 	struct archive_data *dfile;
 	int no;
+	int mixer_no;
 
 	// audio file data
 	SNDFILE *file;
@@ -73,15 +76,34 @@ struct channel {
 	struct fade fade;
 };
 
+struct mixer {
+	sts_mixer_t mixer;
+	sts_mixer_stream_t stream;
+	int voice;
+	atomic_bool muted;
+	float data[CHUNK_SIZE * 2];
+	char *name;
+
+	struct mixer *parent;
+	struct mixer **children;
+	int nr_children;
+};
+
+static struct mixer *master = NULL;
+static struct mixer *mixers = NULL;
+static int nr_mixers = 0;
+
 static SDL_AudioDeviceID audio_device = 0;
-static sts_mixer_t mixer;
 
 /*
  * The SDL2 audio callback.
  */
 static void audio_callback(possibly_unused void *data, Uint8 *stream, int len)
 {
-	sts_mixer_mix_audio(&mixer, stream, len / (sizeof(float) * 2));
+	sts_mixer_mix_audio(&master->mixer, stream, len / (sizeof(float) * 2));
+	if (master->muted) {
+		memset(stream, 0, len);
+	}
 }
 
 /*
@@ -122,7 +144,6 @@ static bool cb_loop(struct channel *ch)
  * Callback to refill the stream's audio data.
  * Called from sys_mixer_mix_audio.
  */
-//static int refill_stream(sts_mixer_sample_t *sample, void *data)
 static int cb_read_frames(struct channel *ch, float *out, sf_count_t frame_count)
 {
 	sf_count_t num_read = 0;
@@ -210,7 +231,7 @@ static int refill_stream(sts_mixer_sample_t *sample, void *data)
 	// set gain for fade
 	if (ch->fade.fading) {
 		float gain = cb_calc_fade(&ch->fade, ch->frame);
-		mixer.voices[ch->voice].gain = gain;
+		mixers[ch->mixer_no].mixer.voices[ch->voice].gain = gain;
 		ch->volume = gain * 100.0;
 
 		if (ch->frame >= ch->fade.end_pos) {
@@ -224,6 +245,17 @@ static int refill_stream(sts_mixer_sample_t *sample, void *data)
 	return r;
 }
 
+static int refill_mixer(sts_mixer_sample_t *sample, void *data)
+{
+	struct mixer *mixer = data;
+	sts_mixer_mix_audio(&mixer->mixer, &mixer->data, CHUNK_SIZE);
+	if (mixer->muted) {
+		memset(mixer->data, 0, sizeof(float) * sample->length);
+	}
+
+	return STS_STREAM_CONTINUE;
+}
+
 int channel_play(struct channel *ch)
 {
 	SDL_LockAudioDevice(audio_device);
@@ -231,7 +263,7 @@ int channel_play(struct channel *ch)
 		SDL_UnlockAudioDevice(audio_device);
 		return 0;
 	}
-	ch->voice = sts_mixer_play_stream(&mixer, &ch->stream, 1.0f);
+	ch->voice = sts_mixer_play_stream(&mixers[ch->mixer_no].mixer, &ch->stream, 1.0f);
 	SDL_UnlockAudioDevice(audio_device);
 	return 1;
 }
@@ -243,7 +275,7 @@ int channel_stop(struct channel *ch)
 		SDL_UnlockAudioDevice(audio_device);
 		return 0;
 	}
-	sts_mixer_stop_voice(&mixer, ch->voice);
+	sts_mixer_stop_voice(&mixers[ch->mixer_no].mixer, ch->voice);
 	ch->voice = -1;
 	SDL_UnlockAudioDevice(audio_device);
 	return 1;
@@ -291,7 +323,7 @@ int channel_fade(struct channel *ch, int time, int volume, bool stop)
 	ch->fade.start_pos = ch->frame;
 	ch->fade.start_volume = (float)ch->volume / 100.0;
 	ch->fade.end_pos = ch->fade.start_pos + ((time * ch->info.samplerate) / 1000);
-	ch->fade.end_volume = min(1.0, max(0.0, (float)volume / 100.0));
+	ch->fade.end_volume = clamp(0.0f, 1.0f, (float)volume / 100.0f);
 	SDL_UnlockAudioDevice(audio_device);
 	return 1;
 }
@@ -395,7 +427,7 @@ static sf_count_t channel_vio_seek(sf_count_t offset, int whence, void *data)
 		ch->offset = ch->dfile->size + offset;
 		break;
 	}
-	ch->offset = min((sf_count_t)ch->dfile->size, max(0, ch->offset));
+	ch->offset = clamp(0, (sf_count_t)ch->dfile->size, ch->offset);
 	return ch->offset;
 }
 
@@ -459,17 +491,28 @@ struct channel *channel_open(enum asset_type type, int no)
 	ch->stream.sample.data = ch->data;
 	ch->voice = -1;
 
-	struct bgi *bgi = bgi_get(no);
-	if (bgi) {
-		ch->volume = min(100, max(0, bgi->volume));
-		ch->loop_start = min(ch->info.frames, max(0, bgi->loop_start));
-		ch->loop_end = min(ch->info.frames, max(0, bgi->loop_end));
-		ch->loop_count = max(0, bgi->loop_count);
-	} else {
+	if (type == ASSET_SOUND) {
+		struct wai *wai = wai_get(no);
 		ch->volume = 100;
 		ch->loop_start = 0;
 		ch->loop_end = ch->info.frames;
-		ch->loop_count = type == ASSET_SOUND ? 1 : 0;
+		ch->loop_count = 1;
+		ch->mixer_no = wai ? wai->channel : 1;
+	} else {
+		struct bgi *bgi = bgi_get(no);
+		if (bgi) {
+			ch->volume = clamp(0, 100, bgi->volume);
+			ch->loop_start = clamp(0, ch->info.frames, bgi->loop_start);
+			ch->loop_end = clamp(0, ch->info.frames, bgi->loop_end);
+			ch->loop_count = max(0, bgi->loop_count);
+			ch->mixer_no = clamp(0, nr_mixers, bgi->channel);
+		} else {
+			ch->volume = 100;
+			ch->loop_start = 0;
+			ch->loop_end = ch->info.frames;
+			ch->loop_count = 0;
+			ch->mixer_no = 0;
+		}
 	}
 	ch->no = no;
 
@@ -490,8 +533,81 @@ void channel_close(struct channel *ch)
 	free(ch);
 }
 
+#define SJIS_MASTER "\x83\x7d\x83\x58\x83\x5e\x81\x5b"
+#define SJIS_VOICE  "\x89\xb9\x90\xba"
+
 void mixer_init(void)
 {
+	// initialize mixer naming
+	if (!config.mixer_nr_channels) {
+		nr_mixers = 3;
+		mixers = xcalloc(nr_mixers, sizeof(struct mixer));
+		mixers[0].name = "Music";
+		mixers[1].name = "Sound";
+		mixers[2].name = "Master";
+	} else {
+		// NOTE: Older games don't have an explicit master channel.
+		//       We add an extra mixer in this case for consistency.
+		int need_master = 1;
+		if (!strcmp(config.mixer_channels[0], "Master") || !strcmp(config.mixer_channels[0], SJIS_MASTER)) {
+			need_master = 0;
+		}
+		nr_mixers = config.mixer_nr_channels + need_master;
+		mixers = xcalloc(nr_mixers, sizeof(struct mixer));
+		for (unsigned i = 0; i < config.mixer_nr_channels; i++) {
+			mixers[i].name = config.mixer_channels[i];
+		}
+		if (need_master) {
+			mixers[nr_mixers-1].name = "Master";
+			master = &mixers[nr_mixers-1];
+		} else {
+			master = &mixers[0];
+		}
+	}
+
+	// initialize mixer hierarchy
+	// NOTE: "Master" and "Voice" are special nodes; all channels following
+	//       them in the list become their children. This appears to be
+	//       hardcoded behavior in system 4.
+	struct mixer *parent = master;
+	for (int i = 0; i < nr_mixers; i++) {
+		if (&mixers[i] == master)
+			continue;
+		mixers[i].parent = parent;
+		parent->children = xrealloc_array(parent->children, parent->nr_children, parent->nr_children+1, sizeof(struct mixer*));
+		parent->children[parent->nr_children++] = &mixers[i];
+		if (!strcmp(mixers[i].name, "Voice") || !strcmp(mixers[i].name, SJIS_VOICE)) {
+			parent = &mixers[i];
+		}
+	}
+
+	// initialize mixers
+	for (int i = 0; i < nr_mixers; i++) {
+		sts_mixer_init(&mixers[i].mixer, 44100, STS_MIXER_SAMPLE_FORMAT_FLOAT);
+		int volume = i < (int)config.mixer_nr_channels ? config.mixer_volumes[i] : config.default_volume;
+		mixers[i].mixer.gain = clamp(0.0f, 1.0f, (float)volume / 100.0f);
+	}
+
+	// initialize mixer streams
+	for (int i = 0; i < nr_mixers; i++) {
+		if (&mixers[i] == master)
+			continue;
+		mixers[i].stream.userdata = &mixers[i];
+		mixers[i].stream.callback = refill_mixer;
+		mixers[i].stream.sample.frequency = 44100;
+		mixers[i].stream.sample.audio_format = STS_MIXER_SAMPLE_FORMAT_FLOAT;
+		mixers[i].stream.sample.length = CHUNK_SIZE * 2;
+		mixers[i].stream.sample.data = mixers[i].data;
+		mixers[i].voice = sts_mixer_play_stream(&mixers[i].parent->mixer, &mixers[i].stream, 1.0f);
+	}
+
+	// read audio metadata
+	if (config.bgi_path)
+		bgi_read(config.bgi_path);
+	if (config.wai_path)
+		wai_load(config.wai_path);
+
+	// initialize SDL audio
 	SDL_AudioSpec have;
 	SDL_AudioSpec want = {
 		.format = AUDIO_F32,
@@ -501,9 +617,53 @@ void mixer_init(void)
 		.callback = audio_callback,
 	};
 	audio_device = SDL_OpenAudioDevice(NULL, 0, &want, &have, 0);
-	sts_mixer_init(&mixer, 44100, STS_MIXER_SAMPLE_FORMAT_FLOAT);
-	if (config.bgi_path)
-		bgi_read(config.bgi_path);
-
 	SDL_PauseAudioDevice(audio_device, 0);
 }
+
+int mixer_get_numof(void)
+{
+	return nr_mixers;
+}
+
+const char *mixer_get_name(int n)
+{
+	if (n < 0 || n >= nr_mixers)
+		return NULL;
+	return mixers[n].name;
+}
+
+int mixer_get_volume(int n, int *volume)
+{
+	if (n < 0 || n >= nr_mixers) {
+		return 0;
+	}
+	SDL_LockAudioDevice(audio_device);
+	*volume = clamp(0, 100, (int)(mixers[n].mixer.gain * 100));
+	SDL_UnlockAudioDevice(audio_device);
+	return 1;
+}
+
+int mixer_set_volume(int n, int volume)
+{
+	if (n < 0 || n >= nr_mixers)
+		return 0;
+	SDL_LockAudioDevice(audio_device);
+	mixers[n].mixer.gain = clamp(0.0f, 1.0f, (float)volume / 100.0f);
+	SDL_UnlockAudioDevice(audio_device);
+	return 1;
+}
+int mixer_get_mute(int n, int *mute)
+{
+	if (n < 0 || n >= nr_mixers)
+		return 0;
+	*mute = mixers[n].muted;
+	return 1;
+}
+
+int mixer_set_mute(int n, int mute)
+{
+	if (n < 0 || n >= nr_mixers)
+		return 0;
+	mixers[n].muted = !!mute;
+	return 1;
+}
diff --git a/src/hll/SystemService.c b/src/hll/SystemService.c
index a4bcdb7..33e2396 100644
--- a/src/hll/SystemService.c
+++ b/src/hll/SystemService.c
@@ -18,56 +18,21 @@
 
 #include "system4/string.h"
 
+#include "mixer.h"
 #include "xsystem4.h"
 #include "hll.h"
 
-static int SystemService_GetMixerNumof(void)
-{
-	return config.mixer_nr_channels;
-}
-
 static int SystemService_GetMixerName(int n, struct string **name)
 {
-	if (n < 0 || (size_t)n >= config.mixer_nr_channels)
+	const char *r = mixer_get_name(n);
+	if (!r)
 		return 0;
-	*name = make_string(config.mixer_channels[n], strlen(config.mixer_channels[n]));
+	*name = make_string(r, strlen(r));
 	return 1;
 }
 
-static int SystemService_GetMixerVolume(int n, int *volume)
-{
-	if (n < 0 || (size_t)n >= config.mixer_nr_channels)
-		return 0;
-	*volume = 100;
-	return 1;
-}
-
-static int SystemService_GetMixerMute(int n, int *mute)
-{
-	if (n < 0 || (size_t)n >= config.mixer_nr_channels)
-		return 0;
-	*mute = 0;
-	return true;
-}
-
 //bool SystemService_SetMixerName(int nNum, string szName);
 
-static int SystemService_SetMixerVolume(int n, possibly_unused int volume)
-{
-	if (n < 0 || (size_t)n >= config.mixer_nr_channels)
-		return 0;
-	// TODO
-	return 1;
-}
-
-static int SystemService_SetMixerMute(int n, possibly_unused int mute)
-{
-	if (n < 0 || (size_t)n >= config.mixer_nr_channels)
-		return 0;
-	// TODO
-	return 1;
-}
-
 //int SystemService_GetGameVersion(void);
 static void SystemService_GetGameName(struct string **game_name)
 {
@@ -163,13 +128,13 @@ static bool SystemService_GetMouseCursorConfig(int type, int *value)
 static void SystemService_RestrainScreensaver(void) { }
 
 HLL_LIBRARY(SystemService,
-	    HLL_EXPORT(GetMixerNumof, SystemService_GetMixerNumof),
+	    HLL_EXPORT(GetMixerNumof, mixer_get_numof),
 	    HLL_EXPORT(GetMixerName, SystemService_GetMixerName),
-	    HLL_EXPORT(GetMixerVolume, SystemService_GetMixerVolume),
-	    HLL_EXPORT(GetMixerMute, SystemService_GetMixerMute),
+	    HLL_EXPORT(GetMixerVolume, mixer_get_volume),
+	    HLL_EXPORT(GetMixerMute, mixer_get_mute),
 	    HLL_TODO_EXPORT(SetMixerName, SystemService_SetMixerName),
-	    HLL_EXPORT(SetMixerVolume, SystemService_SetMixerVolume),
-	    HLL_EXPORT(SetMixerMute, SystemService_SetMixerMute),
+	    HLL_EXPORT(SetMixerVolume, mixer_set_volume),
+	    HLL_EXPORT(SetMixerMute, mixer_set_mute),
 	    HLL_TODO_EXPORT(GetGameVersion, SystemService_GetGameVersion),
 	    HLL_EXPORT(GetGameName, SystemService_GetGameName),
 	    HLL_TODO_EXPORT(AddURLMenu, SystemService_AddURLMenu),
diff --git a/src/system4.c b/src/system4.c
index 76ae709..d81168c 100644
--- a/src/system4.c
+++ b/src/system4.c
@@ -51,6 +51,7 @@ struct config config = {
 	.view_height = 600,
 	.mixer_nr_channels = 0,
 	.mixer_channels = NULL,
+	.default_volume = 100,
 	.joypad = false,
 };
 
@@ -120,6 +121,7 @@ static void read_mixer_channels(struct ini_entry *entry)
 
 	config.mixer_nr_channels = entry->value.list_size;
 	config.mixer_channels = xcalloc(entry->value.list_size, sizeof(char*));
+	config.mixer_volumes = xcalloc(entry->value.list_size, sizeof(int));
 	for (size_t i = 0; i < entry->value.list_size; i++) {
 		if (entry->value.list[i].type == INI_NULL) {
 			WARNING("No name given for VolumeValancer[%d]", i);
@@ -127,6 +129,8 @@ static void read_mixer_channels(struct ini_entry *entry)
 		}
 		if (entry->value.list[i].type == INI_STRING) {
 			config.mixer_channels[i] = strdup(entry->value.list[i].s->text);
+			// XXX: assumes default_volume was already initialized
+			config.mixer_volumes[i] = config.default_volume;
 		} else if (entry->value.list[i].type == INI_LIST) {
 			if (entry->value.list[i].list_size != 2)
 				ERROR("ini value for 'VolumeValancer' is list of wrong size");
@@ -135,6 +139,7 @@ static void read_mixer_channels(struct ini_entry *entry)
 			if (entry->value.list[i].list[1].type != INI_INTEGER)
 				ERROR("ini value for 'VolumeValancer' level is not an integer");
 			config.mixer_channels[i] = strdup(entry->value.list[i].list[0].s->text);
+			config.mixer_volumes[i] = entry->value.list[i].list[1].i;
 		}
 	}
 }
@@ -159,6 +164,8 @@ static void read_config(const char *path)
 			config.view_height = ini_integer(&ini[i]);
 		} else if (!strcmp(ini[i].name->text, "VolumeValancer")) {
 			read_mixer_channels(&ini[i]);
+		} else if (!strcmp(ini[i].name->text, "DefaultVolumeRate")) {
+			config.default_volume = ini_integer(&ini[i]);
 		}
 		ini_free_entry(&ini[i]);
 	}