This suppresses most hiccups. But it also decreases precision to
slightly noticeable levels: about +/-150ms here. For reference, our
audio output core target is +/-40ms.