I actually wrote a short-ass tutorial on that a while back, too. Mostly it's just JSNI, though. http://flax.ie/how-to-use-html5-audio-tag-with-gwt/#more-152
My engine code for the audio component is more or less the same as that, actually. I gave a small, quick explanation on that the other day: http://flax.ie/flax-html5-game-engine-development-diary-part...
</spam>
I found it fairly easy to do. We load the html for each sound object into the page, and then statically play/pause/stop/load them via the tag name. There's a JSON file, which contains a bunch of AudioContainer objects. When they're constructed, they construct their own HTML (so, the audio tag), and loads it into the page. Then we have a static-ish Audio service to make play/pause etc pretty easy. The JSON itself is made in our (as yet hypothetical) map editor. Must get to work on that, actually. ;)
I'm not massively happy with our solution, though. It works, but it's more complex than I feel it should be, considering that the actual implementation of it in js is so easy.
edit: I should point out that the entire point of the engine is so that you don't need Flash, which is why we're using <audio> rather than one of the pre-existing frameworks, like gwt-voices, which we would so totally use, if it didn't use Flash underneath somewhere.