Determinants of the irrelevant speech effect: Changes in spectrum and envelope.

The irrelevant sound effect (ISE) denotes the fact that short-term memory is disrupted while being exposed to sound. The ISE is largest for speech. The presented study investigated the underlying acoustic properties that cause the ISE. Stimuli contained changes in either the spectral content only, the envelope only, or both. For this purpose two experiments were conducted and two vocoding strategies were developed to degrade the spectral content of speech and the envelope independently. The first strategy employed a noise vocoder that was based on perceptual dimensions, analyzing the original utterance into 1, 2, 4, 8, or 24 channels (critical bands) and independently manipulating loudness. The second strategy involved a temporal segmentation of the signal, freezing either spectrum or level for durations ranging from 50 ms to 14 s. In both experiments, changes in envelope alone did not have measurable effects on performance, but the ISE was significantly increased when both the spectral content and the envelope varied. Furthermore, when the envelope changes were uncorrelated with the spectral changes, the effect size was the same as with a constant-loudness envelope. This suggests that the ISE is primarily caused by spectral changes, but concurrent changes in level tend to amplify it.