since we can't assume a 1-to-1 relationship between file and sound, if we start segmenting as in #150 #154

This feels like a more functional approach.
We can add a read method to Soundfile that will return a Sound, but the Sound itself should not be restricted to representing just an entire sound file