diff --git a/Core/apu.c b/Core/apu.c
index 9d78cef..9ef806b 100755
--- a/Core/apu.c
+++ b/Core/apu.c
@@ -62,10 +62,46 @@ static void render(GB_gameboy_t *gb)
     gb->apu_output.cycles_since_render = 0;
     
 
+
+    GB_sample_t filtered_output = gb->apu_output.highpass_mode?
+        (GB_sample_t) {output.left - gb->apu_output.highpass_diff.left,
+                       output.right - gb->apu_output.highpass_diff.right} :
+        output;
+    
+    switch (gb->apu_output.highpass_mode) {
+        case GB_HIGHPASS_OFF:
+            gb->apu_output.highpass_diff = (GB_double_sample_t) {0, 0};
+            break;
+        case GB_HIGHPASS_ACCURATE:
+            gb->apu_output.highpass_diff = (GB_double_sample_t)
+                {output.left - filtered_output.left * gb->apu_output.highpass_rate,
+                    output.right - filtered_output.right * gb->apu_output.highpass_rate};
+            break;
+        case GB_HIGHPASS_REMOVE_DC_OFFSET: {
+            unsigned mask = gb->io_registers[GB_IO_NR51];
+            unsigned left_volume = 0;
+            unsigned right_volume = 0;
+            for (unsigned i = GB_N_CHANNELS; i--;) {
+                if (mask & 1) {
+                    left_volume += (gb->io_registers[GB_IO_NR50] & 7) * CH_STEP * 0xF;
+                }
+                if (mask & 0x10) {
+                    right_volume += ((gb->io_registers[GB_IO_NR50] >> 4) & 7) * CH_STEP * 0xF;
+                }
+                mask >>= 1;
+            }
+            gb->apu_output.highpass_diff = (GB_double_sample_t)
+            {left_volume * (1 - gb->apu_output.highpass_rate) + gb->apu_output.highpass_diff.left * gb->apu_output.highpass_rate,
+                right_volume * (1 - gb->apu_output.highpass_rate) + gb->apu_output.highpass_diff.right * gb->apu_output.highpass_rate};
+
+        }
+            
+    }
+
     while (gb->apu_output.copy_in_progress);
     while (!__sync_bool_compare_and_swap(&gb->apu_output.lock, false, true));
     if (gb->apu_output.buffer_position < gb->apu_output.buffer_size) {
-        gb->apu_output.buffer[gb->apu_output.buffer_position++] = output;
+        gb->apu_output.buffer[gb->apu_output.buffer_position++] = filtered_output;
     }
     gb->apu_output.lock = false;
 }
@@ -712,3 +748,22 @@ size_t GB_apu_get_current_buffer_length(GB_gameboy_t *gb)
 {
     return  gb->apu_output.buffer_position;
 }
+
+void GB_set_sample_rate(GB_gameboy_t *gb, unsigned int sample_rate)
+{
+    if (gb->apu_output.buffer) {
+        free(gb->apu_output.buffer);
+    }
+    gb->apu_output.buffer_size = sample_rate / 25; // 40ms delay
+    gb->apu_output.buffer = malloc(gb->apu_output.buffer_size * sizeof(*gb->apu_output.buffer));
+    gb->apu_output.sample_rate = sample_rate;
+    gb->apu_output.buffer_position = 0;
+    if (sample_rate) {
+        gb->apu_output.highpass_rate = pow(0.999958,  CPU_FREQUENCY / (double)sample_rate);
+    }
+}
+
+void GB_set_highpass_filter_mode(GB_gameboy_t *gb, GB_highpass_mode_t mode)
+{
+    gb->apu_output.highpass_mode = mode;
+}
diff --git a/Core/apu.h b/Core/apu.h
index 06a0a6e..ff70fb3 100644
--- a/Core/apu.h
+++ b/Core/apu.h
@@ -19,6 +19,12 @@ typedef struct
     int16_t right;
 } GB_sample_t;
 
+typedef struct
+{
+    double left;
+    double right;
+} GB_double_sample_t;
+
 enum GB_CHANNELS {
     GB_SQUARE_1,
     GB_SQUARE_2,
@@ -90,6 +96,12 @@ typedef struct
     
 } GB_apu_t;
 
+typedef enum {
+    GB_HIGHPASS_OFF, // Do not apply any filter, keep DC offset
+    GB_HIGHPASS_ACCURATE, // Apply a highpass filter similar to the one used on hardware
+    GB_HIGHPASS_REMOVE_DC_OFFSET, // Remove DC Offset without affecting the waveform
+} GB_highpass_mode_t;
+
 typedef struct {
     unsigned sample_rate;
     
@@ -108,11 +120,16 @@ typedef struct {
     unsigned last_update[GB_N_CHANNELS];
     GB_sample_t current_sample[GB_N_CHANNELS];
     GB_sample_t summed_samples[GB_N_CHANNELS];
+    
+    GB_highpass_mode_t highpass_mode;
+    double highpass_rate;
+    GB_double_sample_t highpass_diff;
 } GB_apu_output_t;
 
 void GB_set_sample_rate(GB_gameboy_t *gb, unsigned int sample_rate);
 void GB_apu_copy_buffer(GB_gameboy_t *gb, GB_sample_t *dest, size_t count);
 size_t GB_apu_get_current_buffer_length(GB_gameboy_t *gb);
+void GB_set_highpass_filter_mode(GB_gameboy_t *gb, GB_highpass_mode_t mode);
 
 #ifdef GB_INTERNAL
 void GB_apu_write(GB_gameboy_t *gb, uint8_t reg, uint8_t value);
diff --git a/Core/gb.c b/Core/gb.c
index 8a2e9e2..923b403 100755
--- a/Core/gb.c
+++ b/Core/gb.c
@@ -386,17 +386,6 @@ void GB_serial_set_data(GB_gameboy_t *gb, uint8_t data)
     gb->io_registers[GB_IO_IF] |= 8;
 }
 
-void GB_set_sample_rate(GB_gameboy_t *gb, unsigned int sample_rate)
-{
-    if (gb->apu_output.buffer) {
-        free(gb->apu_output.buffer);
-    }
-    gb->apu_output.buffer_size = sample_rate / 25; // 40ms delay
-    gb->apu_output.buffer = malloc(gb->apu_output.buffer_size * sizeof(*gb->apu_output.buffer));
-    gb->apu_output.sample_rate = sample_rate;
-    gb->apu_output.buffer_position = 0;
-}
-
 void GB_disconnect_serial(GB_gameboy_t *gb)
 {
     gb->serial_transfer_start_callback = NULL;
@@ -495,7 +484,6 @@ void GB_switch_model_and_reset(GB_gameboy_t *gb, bool is_cgb)
     }
     gb->is_cgb = is_cgb;
     GB_reset(gb);
-
 }
 
 void *GB_get_direct_access(GB_gameboy_t *gb, GB_direct_access_t access, size_t *size, uint16_t *bank)