Comment 53 for bug 193578

I can see the same problem here.
I have got an AMD athlon 64x2 (duo core)@2.2GHz, 2Gio RAM and 2Gio swap.
Motherboard is an Asus M2N DH with an integrated HDA sound chipset from nvidia.
Graphic card is a nvidia geforce 7300 GS.
Ubuntu version is Intrepid 64bits.

The reproducibility is very problematic. You need a heavy page where firefox is slow to scroll in the first place. For example, a page on deviant art (eg. http://brute-ua.deviantart.com/art/Take-a-seat-81523879 ) with the "collect" banner opened (may be available only when you're logged in). Then you have to seize the scrollbar and swing your mouse up and down violently (ie you have to go really top to bottom and back to top again), several times (that can be twice, or 10 times depending on your "luck") and very fast (don't wait for the bottom of the page to display before you're already on top and bottom again, preferably several times). I don't think scrolling with the mouse wheel is sufficient to trigger any sound interruption.

This isn't a normal use of Firefox, but it is needed to reproduce the problem. For a "normal-use", you need to be loading several tabs in Firefox, and switch tab or scoll down rapidly in a long loading page, or maximize-minimize Firefox quickly, or most likely several of these (ie. quickly scrolling down a heavy loading page with several other tabs loading in the background and then double-click on the Firefox icon in the taskbar to minimize and maximize Firefox, all that in less than 1 second or so). Even so, I don't see this bug more than once or twice a day.

I cannot reproduce it on "lighter" pages. A fixed element is recommended, if possible transparent) Even a javascript test such as the Acid3 test (http://acid3.acidtests.org/) or the sunspider test (http://www2.webkit.org/perf/sunspider-0.9/sunspider.html ) doesn't trigger a sound interruption. I guess that owner of faster computers need to find even worse test cases to be able to reproduce it.
Also I should precise that it is "better" to work in full screen (preferably with a big one, 1600x1200 for me, and with F11 fullscreen on) rather than with a smaller Firefox window. If I resize Firefox to use only half of the horizontal space (around 600px) I'm not able to hear any sound interruption.

Running a heavy benchmark such as one in hardinfo doesn't stop sound, nor does sound encoding with SoundConverter (using 2 procs.). To the contrary, I would say it was easier to make firefox interrupt sound when the system isn't doing any other computation. With sound converter working, Firefox scrolls slower but doesn't interrupt the sound anymore.

Stopping compiz and using metacity instead (System > Preferences > Appearance > setting no visual effect) doesn't completely prevent sound interruption. Actually it doesn't even seem to reduce the problem.

I tried to use Alsa instead of pulseaudio as proposed above. This didn't reduce the problem.

I also tried to renice rhythmbox process (with sudo renice 15) but that didn't change anything.

Only enabling crossfading (seemingly) completely stopped sound interruption.

If I look at a terminal with "top" running when the problem occurs, I can see that Xorg is around > 50% CPU and Firefox around 30%.

Finally I don't think it is related to the sound server. If I play in another player such as Totem or Listen, it isn't interrupted. If I play two sounds, one in rhythmbox and one in totem, and then I scroll in Firefox, only rhythmbox's playing is interrupted, and I hear only the sound played in totem.

So I believe that could be a problem of concurrence between Xorg and rhytmbox, triggered by Firefox.