The impact of prior knowledge on perceiving vocal elements in MIDI-converted music

Illusions in which gaps in sensory evidence are filled in using prior knowledge represent a useful avenue for understanding the constructive nature of perception. The Musical Instrument Digital Interface (MIDI) vocals illusion, wherein listeners perceive the presence of vocal elements in a digitally converted audio format with none present, presents a novel opportunity to characterize the role of prior experience in auditory perceptual filling-in.

In two experiments, participants reported the occurrence and duration of either imprecise or precise vocal elements in MIDI-converted audio. To isolate the effect of prior exposure on the emergence of the illusion in each experiment, the participants first listened to 12 MIDI-converted excerpts from a subset of six songs, with one half originally containing vocal elements and the others containing only instrumental tracks. Of the six songs, three were designated as "learned" and were presented in their original format during a subsequent learning block, and the remaining three were only presented in the MIDI format. This block sequence was repeated three times.

An imprecise perceptual illusion emerged regardless of prior exposure to original excerpts and distinguished between excerpts originally containing vocals and those containing only instrumental elements. A more precise illusory percept (words) emerged only for those MIDI stimuli corresponding to the original excerpts presented during the learning blocks.

These findings represent the first investigation of the MIDI vocals illusion and highlight distinct roles of bottom-up sensory features and top-down expectations based on experience in the perceptual filling-in of auditory information.