Spectral and Temporal Features as Estimators of the Irrelevant Speech Effect
Regular paper
Philips Research Laboratories, NL
Wednesday 3 june, 2015, 09:40 - 10:00
0.6 Madrid (49)
Abstract:
The distractive effects on cognitive processes ascribed to the nature of sound
have
been studied in the paradigm of “irrelevant sound,” where test participants
perform cognitive
tasks in the presence of background noise. By comparing the test scores for
different acoustic
stimulus conditions in such experiments, the “irrelevant sound (speech)
effect” (ISE) can be
quantified. The ISE is often explained by the changing state hypothesis: the
distinctive
segmentation of sound tokens; where tokens may be understood as sound segments
that can
be distinguished from each other in temporal and/or spectral characteristics.
A sequence of
sounds consisting of differing tokens produces much more disruption than a
steady-state
sound. The present work investigates the relationship between the features
from both
temporal and spectral domains and the ISE, predicting separately the magnitude
of the effect
with two estimators: The Average Modulation Transfer Function (AMTF) and the
Frequency
Domain Correlation Coefficient (FDCC). The first parameter is a measure for
temporal
variations in a sound, whilst the latter measures spectral variability in the
sounds. Background
stimuli are synthesized from a pulse train in which modified and unmodified
pulses alternate.
In order to manipulate the temporal and spectral features in the stimuli, a
numerical
optimization method was used to generate two sets of background stimuli where
one of the
two descriptors was always kept constant and the other was varied in a
systematic way.
Therefore, stimulus sets used in this study allow the separate estimation of
the role of the two
estimators on cognitive performance in tasks involving serial ordering of
short-term memory
content.