diff --git a/audio/audio.c b/audio/audio.c
index dad17e59b8..4836ab8ca8 100644
--- a/audio/audio.c
+++ b/audio/audio.c
@@ -731,16 +731,21 @@ static size_t audio_pcm_sw_write(SWVoiceOut *sw, void *buf, size_t buf_len)
     hw_free = hw_free > live ? hw_free - live : 0;
     frames_out_max = MIN(dead, hw_free);
     sw_max = st_rate_frames_in(sw->rate, frames_out_max);
-    fe_max = MIN(buf_len / sw->info.bytes_per_frame, sw->resample_buf.size);
+    fe_max = MIN(buf_len / sw->info.bytes_per_frame + sw->resample_buf.pos,
+                 sw->resample_buf.size);
     frames_in_max = MIN(sw_max, fe_max);
 
     if (!frames_in_max) {
         return 0;
     }
 
-    sw->conv(sw->resample_buf.buffer, buf, frames_in_max);
-    if (!sw->hw->pcm_ops->volume_out) {
-        mixeng_volume(sw->resample_buf.buffer, frames_in_max, &sw->vol);
+    if (frames_in_max > sw->resample_buf.pos) {
+        sw->conv(sw->resample_buf.buffer + sw->resample_buf.pos,
+                 buf, frames_in_max - sw->resample_buf.pos);
+        if (!sw->hw->pcm_ops->volume_out) {
+            mixeng_volume(sw->resample_buf.buffer + sw->resample_buf.pos,
+                          frames_in_max - sw->resample_buf.pos, &sw->vol);
+        }
     }
 
     audio_pcm_sw_resample_out(sw, frames_in_max, frames_out_max,
@@ -749,6 +754,22 @@ static size_t audio_pcm_sw_write(SWVoiceOut *sw, void *buf, size_t buf_len)
     sw->total_hw_samples_mixed += total_out;
     sw->empty = sw->total_hw_samples_mixed == 0;
 
+    /*
+     * Upsampling may leave one audio frame in the resample buffer. Decrement
+     * total_in by one if there was a leftover frame from the previous resample
+     * pass in the resample buffer. Increment total_in by one if the current
+     * resample pass left one frame in the resample buffer.
+     */
+    if (frames_in_max - total_in == 1) {
+        /* copy one leftover audio frame to the beginning of the buffer */
+        *sw->resample_buf.buffer = *(sw->resample_buf.buffer + total_in);
+        total_in += 1 - sw->resample_buf.pos;
+        sw->resample_buf.pos = 1;
+    } else if (total_in >= sw->resample_buf.pos) {
+        total_in -= sw->resample_buf.pos;
+        sw->resample_buf.pos = 0;
+    }
+
 #ifdef DEBUG_OUT
     dolog (
         "%s: write size %zu written %zu total mixed %zu\n",
@@ -1155,8 +1176,9 @@ static void audio_run_out (AudioState *s)
                 } else {
                     free = 0;
                 }
-                if (free > 0) {
-                    free = MIN(free, sw->resample_buf.size);
+                if (free > sw->resample_buf.pos) {
+                    free = MIN(free, sw->resample_buf.size)
+                           - sw->resample_buf.pos;
                     sw->callback.fn(sw->callback.opaque,
                                     free * sw->info.bytes_per_frame);
                 }
diff --git a/audio/audio_template.h b/audio/audio_template.h
index a0b653f52c..0d8aab6fad 100644
--- a/audio/audio_template.h
+++ b/audio/audio_template.h
@@ -138,6 +138,12 @@ static int glue (audio_pcm_sw_alloc_resources_, TYPE) (SW *sw)
         return -1;
     }
 
+    /*
+     * Allocate one additional audio frame that is needed for upsampling
+     * if the resample buffer size is small. For large buffer sizes take
+     * care of overflows.
+     */
+    samples = samples < INT_MAX ? samples + 1 : INT_MAX;
     sw->resample_buf.buffer = g_new0(st_sample, samples);
     sw->resample_buf.size = samples;
     sw->resample_buf.pos = 0;