From an e-mail I sent out to audiere-devel:

----
The following is all just guessing based on what I've seen.  As Donald
Norman calls it, "naive physics".  :) But it's useful information
nonetheless...

After doing more testing, it seems that DirectSound doesn't like to
mix very small buffers.  On my machine, when I create and play a
DirectSound buffer smaller than 1000 samples, I get lots of nasty
clicking and popping both before and after the sound itself.  It seems
this is what Enno was running into on Windows XP, just with slightly
larger sound files...  (2500 samples or so.)  When 'streaming' a small
sound, Audiere internally allocates a much larger buffer (two seconds
or so) and feeds data to it.  Since the buffer is bigger, there is no
clicking and popping when playing it.

Strangely, the clicks and pops also occur in the winmm sound driver...
perhaps because winmm only writes audio when it is available, so a
small waveOutWrite is bad...  (?)  I'll have to do more investigation
here.

Anyway, I'm thinking of adding a device parameter to the directsound
device so that the user can specify the minimum buffer size.  Any
buffer smaller than that size will be enlarged (with silence) before
being passed to DirectSound.  It seems to me that this would solve the
problem...  Any thoughts?  What should the default size of the buffer
be?
----

Some URLs I found:

http://sound.media.mit.edu/mpeg4/saol-users/0250.html
http://msdn.microsoft.com/library/default.asp?url=/library/en-us/directx9_c/directx/htm/usingbuffersefficiently.asp


The best long-term solution is to implement a pure software mixer
built upon a low-latency audio output system such as PortAudio, ASIO,
or DirectSound.  The mixer must be much higher quality and more
optimized than the one in Audiere now.  It should do cubic resampling
and use MMX or SSE to mix.