My personal theory is that stuttering is just the "bad" end of a scale of fluency, which can be likened to RAM memory in a computer. Stutterers have less of a "buffer" in which to prepare the sequence of words, and we start stuttering when we try to parse words that we haven't had the time to buffer. In singing, however, the cadence is pre\-defined by the melody of the song, and we only have to worry about following the tune and remembering the lyrics; which does not strain the buffer as much as speaking, where we have to "wing" both the content and the inflection to match the context of the conversation.