diff --git a/miniaudio.h b/miniaudio.h
index 7f0b9784..1ef28232 100644
--- a/miniaudio.h
+++ b/miniaudio.h
@@ -7962,7 +7962,7 @@ struct ma_device
     ma_bool8 noDisableDenormals;
     ma_bool8 noFixedSizedCallback;
     ma_atomic_float masterVolumeFactor;             /* Linear 0..1. Can be read and written simultaneously by different threads. Must be used atomically. */
-    ma_duplex_rb duplexRB;                          /* Intermediary buffer for duplex devices on asynchronous backends. */
+    ma_duplex_rb* pDuplexRB;                        /* Intermediary buffer for duplex devices. Allocated on the heap to ensure it is aligned to MA_CACHE_LINE_SIZE. Will be null for non-duplex devices. */
     struct
     {
         ma_resample_algorithm algorithm;
@@ -49202,7 +49202,15 @@ MA_API ma_result ma_device_init(ma_context* pContext, const ma_device_config* pC
     after ma_device_post_init().
     */
     if (pConfig->deviceType == ma_device_type_duplex) {
-        result = ma_duplex_rb_init(pDevice->capture.format, pDevice->capture.channels, pDevice->sampleRate, pDevice->capture.internalSampleRate, pDevice->capture.internalPeriodSizeInFrames, ma_device_get_allocation_callbacks(pDevice), &pDevice->duplexRB);
+        /* TODO: Allocate one block of memory big enough for the ma_duplex_rb struct *and* the internal buffer. Then pass in the pre-allocated buffer to the ring buffer init routine. */
+
+        pDevice->pDuplexRB = (ma_duplex_rb*)ma_aligned_malloc(sizeof(ma_duplex_rb), MA_CACHE_LINE_SIZE, ma_device_get_allocation_callbacks(pDevice));
+        if (pDevice->pDuplexRB == NULL) {
+            ma_device_uninit(pDevice);
+            return MA_OUT_OF_MEMORY;
+        }
+
+        result = ma_duplex_rb_init(pDevice->capture.format, pDevice->capture.channels, pDevice->sampleRate, pDevice->capture.internalSampleRate, pDevice->capture.internalPeriodSizeInFrames, ma_device_get_allocation_callbacks(pDevice), pDevice->pDuplexRB);
         if (result != MA_SUCCESS) {
             ma_device_uninit(pDevice);
             return result;
@@ -49415,7 +49423,8 @@ MA_API void ma_device_uninit(ma_device* pDevice)
     }
 
     if (pDevice->type == ma_device_type_duplex) {
-        ma_duplex_rb_uninit(&pDevice->duplexRB);
+        ma_duplex_rb_uninit(pDevice->pDuplexRB);
+        ma_aligned_free(pDevice->pDuplexRB, ma_device_get_allocation_callbacks(pDevice));
     }
 
     if (pDevice->type == ma_device_type_capture || pDevice->type == ma_device_type_duplex || pDevice->type == ma_device_type_loopback) {
@@ -49993,12 +50002,14 @@ MA_API ma_result ma_device_handle_backend_data_callback(ma_device* pDevice, void
     }
 
     if (pDevice->type == ma_device_type_duplex) {
+        MA_ASSERT(pDevice->pDuplexRB != NULL);
+
         if (pInput != NULL) {
-            ma_device__handle_duplex_callback_capture(pDevice, frameCount, pInput, &pDevice->duplexRB.rb);
+            ma_device__handle_duplex_callback_capture(pDevice, frameCount, pInput, &pDevice->pDuplexRB->rb);
         }
 
         if (pOutput != NULL) {
-            ma_device__handle_duplex_callback_playback(pDevice, frameCount, pOutput, &pDevice->duplexRB.rb);
+            ma_device__handle_duplex_callback_playback(pDevice, frameCount, pOutput, &pDevice->pDuplexRB->rb);
         }
     } else {
         if (pDevice->type == ma_device_type_capture || pDevice->type == ma_device_type_loopback) {