diff --git a/src/NDS.cpp b/src/NDS.cpp
index d8f346e8..623f025d 100644
--- a/src/NDS.cpp
+++ b/src/NDS.cpp
@@ -25,6 +25,7 @@
 #include "DMA.h"
 #include "FIFO.h"
 #include "GPU.h"
+#include "SPU.h"
 #include "SPI.h"
 #include "RTC.h"
 #include "Wifi.h"
@@ -37,11 +38,6 @@ namespace NDS
 // * stick all the variables in a big structure?
 //   would make it easier to deal with savestates
 
-/*SchedEvent SchedBuffer[SCHED_BUF_LEN];
-SchedEvent* SchedQueue;
-
-bool NeedReschedule;*/
-
 ARM* ARM9;
 ARM* ARM7;
 
@@ -108,8 +104,6 @@ u32 SqrtRes;
 
 u32 KeyInput;
 
-u16 _soundbias; // temp
-
 bool Running;
 
 
@@ -132,6 +126,7 @@ bool Init()
 
     if (!NDSCart::Init()) return false;
     if (!GPU::Init()) return false;
+    if (!SPU::Init()) return false;
     if (!SPI::Init()) return false;
     if (!RTC::Init()) return false;
 
@@ -151,6 +146,7 @@ void DeInit()
 
     NDSCart::DeInit();
     GPU::DeInit();
+    SPU::DeInit();
     SPI::DeInit();
     RTC::DeInit();
 }
@@ -300,14 +296,6 @@ void Reset()
     for (i = 0; i < 8; i++) DMAs[i]->Reset();
     memset(DMA9Fill, 0, 4*4);
 
-    NDSCart::Reset();
-    GPU::Reset();
-    SPI::Reset();
-    RTC::Reset();
-    Wifi::Reset();
-
-   // memset(SchedBuffer, 0, sizeof(SchedEvent)*SCHED_BUF_LEN);
-   // SchedQueue = NULL;
     memset(SchedList, 0, sizeof(SchedList));
     SchedListMask = 0;
 
@@ -319,7 +307,12 @@ void Reset()
 
     KeyInput = 0x007F03FF;
 
-    _soundbias = 0;
+    NDSCart::Reset();
+    GPU::Reset();
+    SPU::Reset();
+    SPI::Reset();
+    RTC::Reset();
+    Wifi::Reset();
 }
 
 void LoadROM(const char* path, bool direct)
@@ -1629,7 +1622,6 @@ void ARM9IOWrite16(u32 addr, u16 val)
         {
             SetIRQ(1, IRQ_IPCSync);
         }
-        //CompensateARM7();
         return;
 
     case 0x04000184:
@@ -1901,8 +1893,7 @@ u8 ARM7IORead8(u32 addr)
 
     if (addr >= 0x04000400 && addr < 0x04000520)
     {
-        // sound I/O
-        return 0;
+        return SPU::Read8(addr);
     }
 
     printf("unknown ARM7 IO read8 %08X\n", addr);
@@ -1972,14 +1963,11 @@ u16 ARM7IORead16(u32 addr)
     case 0x04000300: return PostFlag7;
     case 0x04000304: return PowerControl7;
     case 0x04000308: return ARM7BIOSProt;
-
-    case 0x04000504: return _soundbias;
     }
 
     if (addr >= 0x04000400 && addr < 0x04000520)
     {
-        // sound I/O
-        return 0;
+        return SPU::Read16(addr);
     }
 
     printf("unknown ARM7 IO read16 %08X %08X\n", addr, ARM9->R[15]);
@@ -2057,8 +2045,7 @@ u32 ARM7IORead32(u32 addr)
 
     if (addr >= 0x04000400 && addr < 0x04000520)
     {
-        // sound I/O
-        return 0;
+        return SPU::Read32(addr);
     }
 
     printf("unknown ARM7 IO read32 %08X\n", addr);
@@ -2116,7 +2103,7 @@ void ARM7IOWrite8(u32 addr, u8 val)
 
     if (addr >= 0x04000400 && addr < 0x04000520)
     {
-        // sound I/O
+        SPU::Write8(addr, val);
         return;
     }
 
@@ -2228,15 +2215,11 @@ void ARM7IOWrite16(u32 addr, u16 val)
         if (ARM7BIOSProt == 0)
             ARM7BIOSProt = val;
         return;
-
-    case 0x04000504: // removeme
-        _soundbias = val & 0x3FF;
-        return;
     }
 
     if (addr >= 0x04000400 && addr < 0x04000520)
     {
-        // sound I/O
+        SPU::Write16(addr, val);
         return;
     }
 
@@ -2326,7 +2309,7 @@ void ARM7IOWrite32(u32 addr, u32 val)
 
     if (addr >= 0x04000400 && addr < 0x04000520)
     {
-        // sound I/O
+        SPU::Write32(addr, val);
         return;
     }
 
diff --git a/src/NDS.h b/src/NDS.h
index 4fec1174..583e12c1 100644
--- a/src/NDS.h
+++ b/src/NDS.h
@@ -24,30 +24,11 @@
 namespace NDS
 {
 
-/*#define SCHED_BUF_LEN 64
-
-typedef struct _SchedEvent
-{
-    u32 Delay;
-    void (*Func)(u32);
-    u32 Param;
-    struct _SchedEvent* PrevEvent;
-    struct _SchedEvent*  NextEvent;
-
-} SchedEvent;*/
-
 enum
 {
     Event_LCD = 0,
 
-    /*Event_Timer9_0,
-    Event_Timer9_1,
-    Event_Timer9_2,
-    Event_Timer9_3,
-    Event_Timer7_0,
-    Event_Timer7_1,
-    Event_Timer7_2,
-    Event_Timer7_3,*/
+    Event_SPU,
 
     Event_MAX
 };
@@ -95,7 +76,6 @@ typedef struct
     u16 Cnt;
     u32 Counter;
     u32 CycleShift;
-    //SchedEvent* Event;
 
 } Timer;
 
@@ -127,15 +107,9 @@ void ReleaseKey(u32 key);
 void TouchScreen(u16 x, u16 y);
 void ReleaseScreen();
 
-/*SchedEvent* ScheduleEvent(s32 Delay, void (*Func)(u32), u32 Param);
-void CancelEvent(SchedEvent* event);
-void RunEvents(s32 cycles);*/
 void ScheduleEvent(u32 id, bool periodic, s32 delay, void (*func)(u32), u32 param);
 void CancelEvent(u32 id);
 
-// DO NOT CALL FROM ARM7!!
-void CompensateARM7();
-
 void debug(u32 p);
 
 void Halt();
diff --git a/src/SPU.cpp b/src/SPU.cpp
index c7a769ee..440d39d9 100644
--- a/src/SPU.cpp
+++ b/src/SPU.cpp
@@ -25,38 +25,269 @@
 namespace SPU
 {
 
-//SDL_AudioDeviceID device;
-//
+const u32 OutputBufferSize = 2*1024;
+s16 OutputBuffer[2 * OutputBufferSize];
+u32 OutputReadOffset;
+u32 OutputWriteOffset;
+
+
+u16 Cnt;
+u8 MasterVolume;
+u16 Bias;
+
+Channel* Channels[16];
 
 
 bool Init()
 {
-    /*SDL_AudioSpec whatIwant, whatIget;
-
-    memset(&whatIwant, 0, sizeof(SDL_AudioSpec));
-    whatIwant.freq = 32824; // 32823.6328125
-    whatIwant.format = AUDIO_S16LSB;
-    whatIwant.channels = 2;
-    whatIwant.samples = 2048;
-    whatIwant.callback = zorp;
-    device = SDL_OpenAudioDevice(NULL, 0, &whatIwant, &whatIget, 0);
-    if (!device)
-    {
-        printf("Audio init failed: %s\n", SDL_GetError());
-        return false;
-    }*/
+    for (int i = 0; i < 16; i++)
+        Channels[i] = new Channel(i);
 
     return true;
 }
 
 void DeInit()
 {
-    //if (device) SDL_CloseAudioDevice(device);
+    for (int i = 0; i < 16; i++)
+        delete Channels[i];
 }
 
 void Reset()
+{
+    memset(OutputBuffer, 0, 2*OutputBufferSize*2);
+    OutputReadOffset = 0;
+    OutputWriteOffset = 0;
+
+    Cnt = 0;
+    MasterVolume = 0;
+    Bias = 0;
+
+    for (int i = 0; i < 16; i++)
+        Channels[i]->Reset();
+
+    NDS::ScheduleEvent(NDS::Event_SPU, true, 1024*16, Mix, 16);
+}
+
+
+Channel::Channel(u32 num)
 {
     //
 }
 
+Channel::~Channel()
+{
+    //
+}
+
+void Channel::Reset()
+{
+    SetCnt(0);
+    SrcAddr = 0;
+    TimerReload = 0;
+    LoopPos = 0;
+    Length = 0;
+}
+
+
+void Mix(u32 samples)
+{
+    //
+
+    NDS::ScheduleEvent(NDS::Event_SPU, true, 1024*16, Mix, 16);
+}
+
+
+void ReadOutput(s16* data, int samples)
+{
+    for (int i = 0; i < samples; i++)
+    {
+        *data++ = OutputBuffer[OutputReadOffset];
+        *data++ = OutputBuffer[OutputReadOffset + 1];
+
+        if (OutputReadOffset != OutputWriteOffset)
+        {
+            OutputReadOffset += 2;
+            OutputReadOffset &= ((2*OutputBufferSize)-1);
+        }
+    }
+}
+
+
+u8 Read8(u32 addr)
+{
+    if (addr < 0x04000500)
+    {
+        Channel* chan = Channels[(addr >> 4) & 0xF];
+
+        switch (addr & 0xF)
+        {
+        case 0x0: return chan->Cnt & 0xFF;
+        case 0x1: return (chan->Cnt >> 8) & 0xFF;
+        case 0x2: return (chan->Cnt >> 16) & 0xFF;
+        case 0x3: return chan->Cnt >> 24;
+        }
+    }
+    else
+    {
+        switch (addr)
+        {
+        case 0x04000500: return Cnt & 0x7F;
+        case 0x04000501: return Cnt >> 8;
+        }
+    }
+
+    printf("unknown SPU read8 %08X\n", addr);
+    return 0;
+}
+
+u16 Read16(u32 addr)
+{
+    if (addr < 0x04000500)
+    {
+        Channel* chan = Channels[(addr >> 4) & 0xF];
+
+        switch (addr & 0xF)
+        {
+        case 0x0: return chan->Cnt & 0xFFFF;
+        case 0x2: return chan->Cnt >> 16;
+        }
+    }
+    else
+    {
+        switch (addr)
+        {
+        case 0x04000500: return Cnt;
+        case 0x04000504: return Bias;
+        }
+    }
+
+    printf("unknown SPU read16 %08X\n", addr);
+    return 0;
+}
+
+u32 Read32(u32 addr)
+{
+    if (addr < 0x04000500)
+    {
+        Channel* chan = Channels[(addr >> 4) & 0xF];
+
+        switch (addr & 0xF)
+        {
+        case 0x0: return chan->Cnt;
+        }
+    }
+    else
+    {
+        switch (addr)
+        {
+        case 0x04000500: return Cnt;
+        case 0x04000504: return Bias;
+        }
+    }
+
+    printf("unknown SPU read32 %08X\n", addr);
+    return 0;
+}
+
+void Write8(u32 addr, u8 val)
+{
+    if (addr < 0x04000500)
+    {
+        Channel* chan = Channels[(addr >> 4) & 0xF];
+
+        switch (addr & 0xF)
+        {
+        case 0x0: chan->SetCnt((chan->Cnt & 0xFFFFFF00) | val); return;
+        case 0x1: chan->SetCnt((chan->Cnt & 0xFFFF00FF) | (val << 8)); return;
+        case 0x2: chan->SetCnt((chan->Cnt & 0xFF00FFFF) | (val << 16)); return;
+        case 0x3: chan->SetCnt((chan->Cnt & 0x00FFFFFF) | (val << 24)); return;
+        }
+    }
+    else
+    {
+        switch (addr)
+        {
+        case 0x04000500:
+            Cnt = (Cnt & 0xBF00) | (val & 0x7F);
+            MasterVolume = Cnt & 0x7F;
+            if (MasterVolume == 127) MasterVolume++;
+            return;
+        case 0x04000501:
+            Cnt = (Cnt & 0x007F) | ((val & 0xBF) << 8);
+            return;
+        }
+    }
+
+    printf("unknown SPU write8 %08X %02X\n", addr, val);
+}
+
+void Write16(u32 addr, u16 val)
+{
+    if (addr < 0x04000500)
+    {
+        Channel* chan = Channels[(addr >> 4) & 0xF];
+
+        switch (addr & 0xF)
+        {
+        case 0x0: chan->SetCnt((chan->Cnt & 0xFFFF0000) | val); return;
+        case 0x2: chan->SetCnt((chan->Cnt & 0x0000FFFF) | (val << 16)); return;
+        case 0x8: chan->SetTimerReload(val); return;
+        case 0xA: chan->SetLoopPos(val); return;
+        }
+    }
+    else
+    {
+        switch (addr)
+        {
+        case 0x04000500:
+            Cnt = val & 0xBF7F;
+            MasterVolume = Cnt & 0x7F;
+            if (MasterVolume == 127) MasterVolume++;
+            return;
+
+        case 0x04000504:
+            Bias = val & 0x3FF;
+            return;
+        }
+    }
+
+    printf("unknown SPU write16 %08X %04X\n", addr, val);
+}
+
+void Write32(u32 addr, u32 val)
+{
+    if (addr < 0x04000500)
+    {
+        Channel* chan = Channels[(addr >> 4) & 0xF];
+
+        switch (addr & 0xF)
+        {
+        case 0x0: chan->SetCnt(val); return;
+        case 0x4: chan->SetSrcAddr(val); return;
+        case 0x8:
+            chan->SetTimerReload(val & 0xFFFF);
+            chan->SetLoopPos(val >> 16);
+            return;
+        case 0xC: chan->SetLength(val); return;
+        }
+    }
+    else
+    {
+        switch (addr)
+        {
+        case 0x04000500:
+            Cnt = val & 0xBF7F;
+            MasterVolume = Cnt & 0x7F;
+            if (MasterVolume == 127) MasterVolume++;
+            return;
+
+        case 0x04000504:
+            Bias = val & 0x3FF;
+            return;
+        }
+    }
+
+    printf("unknown SPU write32 %08X %08X\n", addr, val);
+}
+
 }
diff --git a/src/SPU.h b/src/SPU.h
index a66b98d8..5b762bd3 100644
--- a/src/SPU.h
+++ b/src/SPU.h
@@ -26,6 +26,41 @@ bool Init();
 void DeInit();
 void Reset();
 
+void Mix(u32 samples);
+
+void ReadOutput(s16* data, int samples);
+
+u8 Read8(u32 addr);
+u16 Read16(u32 addr);
+u32 Read32(u32 addr);
+void Write8(u32 addr, u8 val);
+void Write16(u32 addr, u16 val);
+void Write32(u32 addr, u32 val);
+
+class Channel
+{
+public:
+    Channel(u32 num);
+    ~Channel();
+    void Reset();
+
+    u32 Cnt;
+    u32 SrcAddr;
+    u16 TimerReload;
+    u32 LoopPos;
+    u32 Length;
+
+    void SetCnt(u32 val)
+    {
+        Cnt = val & 0xFF7F837F;
+    }
+
+    void SetSrcAddr(u32 val) { SrcAddr = val & 0x07FFFFFF; }
+    void SetTimerReload(u32 val) { TimerReload = val & 0xFFFF; }
+    void SetLoopPos(u32 val) { LoopPos = (val & 0xFFFF) << 2; }
+    void SetLength(u32 val) { Length = (val & 0x001FFFFF) << 2; }
+};
+
 }
 
 #endif // SPU_H
diff --git a/src/wx/main.cpp b/src/wx/main.cpp
index 85568a4f..f3d8ef82 100644
--- a/src/wx/main.cpp
+++ b/src/wx/main.cpp
@@ -22,6 +22,7 @@
 #include "../Config.h"
 #include "../NDS.h"
 #include "../GPU.h"
+#include "../SPU.h"
 
 #include "InputConfig.h"
 #include "EmuConfig.h"
@@ -86,7 +87,7 @@ bool wxApp_melonDS::OnInit()
     printf("melonDS " MELONDS_VERSION "\n" MELONDS_URL "\n");
 
     Config::Load();
-    
+
     emuthread = new EmuThread();
     if (emuthread->Run() != wxTHREAD_NO_ERROR)
     {
@@ -97,7 +98,7 @@ bool wxApp_melonDS::OnInit()
 
     MainFrame* melon = new MainFrame();
     melon->Show(true);
-    
+
     melon->emuthread = emuthread;
     emuthread->parent = melon;
 
@@ -108,7 +109,7 @@ int wxApp_melonDS::OnExit()
 {
     emuthread->Wait();
     delete emuthread;
-    
+
     return wxApp::OnExit();
 }
 
@@ -169,7 +170,7 @@ void MainFrame::OnClose(wxCloseEvent& event)
 {
     emuthread->EmuPause();
     emuthread->EmuExit();
-    
+
     NDS::DeInit();
 
     if (joy)
@@ -313,6 +314,11 @@ EmuThread::~EmuThread()
 {
 }
 
+static void AudioCallback(void* data, Uint8* stream, int len)
+{
+    SPU::ReadOutput((s16*)stream, len>>2);
+}
+
 wxThread::ExitCode EmuThread::Entry()
 {
     emustatus = 3;
@@ -344,6 +350,23 @@ wxThread::ExitCode EmuThread::Entry()
     botdst.x = 0; botdst.y = 192;
     botdst.w = 256; botdst.h = 192;
 
+    SDL_AudioSpec whatIwant, whatIget;
+    memset(&whatIwant, 0, sizeof(SDL_AudioSpec));
+    whatIwant.freq = 32824; // 32823.6328125
+    whatIwant.format = AUDIO_S16LSB;
+    whatIwant.channels = 2;
+    whatIwant.samples = 1024;
+    whatIwant.callback = AudioCallback;
+    audio = SDL_OpenAudioDevice(NULL, 0, &whatIwant, &whatIget, 0);
+    if (!audio)
+    {
+        printf("Audio init failed: %s\n", SDL_GetError());
+    }
+    else
+    {
+        SDL_PauseAudioDevice(audio, 0);
+    }
+
     Touching = false;
     axismask = 0;
 
@@ -430,9 +453,11 @@ wxThread::ExitCode EmuThread::Entry()
             emupaused = true;
         }
     }
-    
+
     emupaused = true;
 
+    if (audio) SDL_CloseAudioDevice(audio);
+
     SDL_DestroyTexture(sdltex);
     SDL_DestroyRenderer(sdlrend);
     SDL_DestroyWindow(sdlwin);
@@ -462,7 +487,7 @@ void EmuThread::ProcessEvents()
                 {
                     int w = evt.window.data1;
                     int h = evt.window.data2;
-                    
+
                     // SDL_SetWindowMinimumSize() doesn't seem to work on Linux. oh well
                     if ((w < 256) || (h < 384))
                     {
@@ -514,7 +539,7 @@ void EmuThread::ProcessEvents()
                 {
                     Touching = true;
                     NDS::PressKey(16+6);
-                    
+
                     int mx, my;
                     SDL_GetGlobalMouseState(&mx, &my);
                     txoffset = mx - evt.button.x;
diff --git a/src/wx/main.h b/src/wx/main.h
index 851a0612..0219ff73 100644
--- a/src/wx/main.h
+++ b/src/wx/main.h
@@ -46,7 +46,7 @@ class wxApp_melonDS : public wxApp
 public:
     virtual bool OnInit();
     virtual int OnExit();
-    
+
     EmuThread* emuthread;
 };
 
@@ -91,7 +91,7 @@ public:
 
     bool EmuIsRunning() { return (emustatus == 1) || (emustatus == 2); }
     bool EmuIsPaused() { return (emustatus == 2) && emupaused; }
-    
+
     MainFrame* parent;
 
 protected:
@@ -105,6 +105,8 @@ protected:
     SDL_Rect topsrc, topdst;
     SDL_Rect botsrc, botdst;
 
+    SDL_AudioDeviceID audio;
+
     bool Touching;
     int txoffset, tyoffset;