SMIL seems to be a formatting language...in which certain sound components can be defined and embedded. It would be used for complex multimedia presentations.

If one wanted to extract a sound portion AS an MP3, it might be that the only way is to play an SMIL presentation and use some program that will record anything passing thru the sound card (there are several)...then convert it to MP3. It is 'probably' easier to find the material already in MP3 format.