Unfortunately, these messages are missing the equivalent of a "camera_info" topic to describe the data (there was at one point a discussion about fixing this, but there are very few people doing audio in ROS right now).
The message itself doesn't constrain you to a particular set of parameters -- however, if you want to use the tools in the audio_common stack, then that sets the parameters. For instance, the audio_capture node will encode the data as 192kbit/s MP3, the number of samples per message is variable I believe.
For the most part, people only use these messages with the audio_capture/audio_play nodes, and so it hasn't been much of a problem. If you are uninterested in using either of these nodes, you can actually put any form of data into the message, although it may be wiser to just create a new message type for your particular application.