If you were setting this up in a lab to test you'd be playing the same song on repeat trying to recreate the circumstances and effect, possibly at different volumes and playback speeds and other things until you isolated a particular portion that caused the crash. Then you'd play back that portion. And all of this would be at a level people could hear. So you'd be hearing the same song over and over in an otherwise quiet lab. Even if you liked the song, you'd hate it by the end of the day (or spend the day wearing earplugs).
One of the reasons why Cortana was deprecated was the reports of beleaguered IT professionals having to listen to dozens of chattering Cortanas as they set up a fleet of new corporate laptops.
