diff --git a/src/cubeb_aaudio.cpp b/src/cubeb_aaudio.cpp
--- a/src/cubeb_aaudio.cpp
+++ b/src/cubeb_aaudio.cpp
@@ -1328,18 +1328,18 @@ aaudio_stream_init_impl(cubeb_stream * s
 
     int rate = WRAP(AAudioStream_getSampleRate)(stm->ostream);
     int32_t output_burst_frames =
         WRAP(AAudioStream_getFramesPerBurst)(stm->ostream);
     // 3 times the burst size seems fairly robust, use it as minimum.
     int32_t output_buffer_size_frames = 3 * output_burst_frames;
     if (stm->latency_frames > POWERSAVE_LATENCY_FRAMES_THRESHOLD) {
       // FramesPerBurst is large in power saving mode, reduce the buffer size to
-      // 1 burst.
-      output_buffer_size_frames = output_burst_frames;
+      // 2 bursts.
+      output_buffer_size_frames = 2 * output_burst_frames;
     }
     // Make output buffer size a function of the requested latency so clients
     // can adapt to their use case.
     output_buffer_size_frames =
         std::max(output_buffer_size_frames,
                  static_cast<int32_t>(stm->latency_frames / 2));
     int32_t output_final_buffer_size_frames =
         WRAP(AAudioStream_setBufferSizeInFrames)(stm->ostream,