Many developers, especially those creating software demonstrations using tools such as Captivate, often create extemporaneous narration as they perform the software demonstration.
Some projects I have had involved using subject matter experts as narrators, who insisted on speaking of the cuff instead of scripting out their comments. These were usually SMEs who had done stand up lectures of their topics, and were confident that they could save effort by reproducing their classroom narrative as the basis for the narration of a elearning version of the program.
Neither of these approaches are optimal.
Voice content created off the top of the SME’s head often either require extensive editing to produce a focused and efficient narrative, or are later abandoned in favor of creating a script (often based on the recorded narrative, as in “…just create a script on what I recorded, and just tighten it up a bit…”).
I would strongly recommend discouraging your SMEs or instructors to attempt to do an off-the-cuff performance as the basis for the narration for an elearning program (though you may have to convince them via a test recording session).