Hi jkl,
What's most puzzling is that this is happening on different computers and in different conditions.
Psychboiler's suggestion is a good one, and double checking your questionnaires and experiments for proper structure is definitely the place to start
If you are intending to record a voice response and that item is present, it might also be helpful to see if there is anything about the 20 participants who do not have voice data but should have in common. Perhaps they are all giving a response or a combination or responses that is causing there to be an error in how the questionnaire/experiment is being administered. We can hope that what appears random will actually have some pattern emerge.