Steady-state responses to concurrent melodies: source distribution, top-down, and bottom-up attention

Abstract

Humans can direct attentional resources to a single sound occurring simultaneously among others to extract the most behaviourally relevant information present. To investigate this cognitive phenomenon in a precise manner, we used frequency-tagging to separate neural auditory steady-state responses (ASSRs) that can be traced back to each auditory stimulus, from the neural mix elicited by multiple simultaneous sounds. Using a mixture of 2 frequency-tagged melody streams, we instructed participants to selectively attend to one stream or the other while following the development of the pitch contour. Bottom-up attention towards either stream was also manipulated with salient changes in pitch. Distributed source analyses of magnetoencephalography measurements showed that the effect of ASSR enhancement from top-down driven attention was strongest at the left frontal cortex, while that of bottom-up driven attention was dominant at the right temporal cortex. Furthermore, the degree of ASSR suppression from simultaneous stimuli varied across cortical lobes and hemisphere. The ASSR source distribution changes from temporal-dominance during single-stream perception, to proportionally more activity in the frontal and centro-parietal cortical regions when listening to simultaneous streams. These findings are a step forward to studying cognition in more complex and naturalistic soundscapes using frequency-tagging.

Keywords: ASSR; MEG; cocktail party; music; simultaneous.

Fig. 1

(Top row) Sample block of melody in the overlapping MDT task. Participants listened to 2 overlapping melody streams while attending to either the Low voice or High voice following a cue. When the melody stopped, participants reported the last direction of pitch change for the attended melody stream (i.e. falling, rising or constant pitch). The respective f_c (pitch) range and f_m of each stream are indicated above. (Bottom row) Color matrix describing how top-down (TD) and bottom-up (BU) attention was studied using the overlapping MDT paradigm. For the top-down driven attended voice, the tone onset (pitch change) draws bottom-up selective attention towards it in the first 1 s (green). At the second half of that attended tone, only top-down attention is still directed to the same voice (blue), while the simultaneous tone onset in the unattended voice generates a bottom-up pull on selective attention (yellow) in conflict with top-down attention. In the second half of the tone in the unattended voice, neither top-down nor bottom-up factors draw attention to the tone (red).

Fig. 2

(Left) Grand average nonoverlapping (—) and overlapping (− −) ASSR power spectra during low tone (blue) and high tone (red) onset. Log-transformed ASSR power at f_m (low: 39 Hz; high: 43 Hz) between nonoverlapping and overlapping conditions was significantly different (P < 0.001) when averaged across all MEG sensors. The ASSR decreased to approximately a third of its nonoverlapping power upon the addition of another simultaneous voice as illustrated by downward arrows above. (Right) Grand average topographic plot of MEG gradiometers at f_m across all 26 subjects. Units are converted to T². While the example above depicts the MEG sensor activity pattern for the overlapping High voice at f_m = 43 Hz, similar patterns of activity were observed for the Low voice at f_m = 39 Hz, and for the nonoverlapping voices.

Fig. 3

Effect of ASSR power suppression due to simultaneous voices. ANOVA results reported significant main effects of (a) hemisphere (P = 0.020), (b) lobe (P < 0.001), and (c) voice (P = 0.013), but no significant interaction between any of these factors. Post-hoc Tukey tests revealed that all possible combinations of the 3 lobe levels were significantly different (P < 0.001, corrected for all combinations as indicated by the asterisks^*** above), with the temporal lobe experiencing the most suppression from simultaneous sources followed by parietal then frontal. The corresponding ASSR2/ASSR1 ratios, displayed in parentheses, were calculated directly from each respective mean lg(ASSR2/ASSR1) value right above. Vertical bars denote 0.95 confidence intervals.

Fig. 4

Demarcation of the areas that define each of the frontal (green), temporal (blue), and parietal (magenta) lobes used for ANOVA analysis of the ASSR suppression due to simultaneous voices. For visualization, these demarcations are superimposed over cortical maps of t-scores (across-subject) illustrating the power difference between ASSR1 and ASSR2 at vertex-level. Orientation views from left to right: right lateral, left lateral, frontal, top.

Fig. 5

Difference in the distribution of ASSR sources generated by overlapping and nonoverlapping voices across the frontal, temporal, and parietal cortices. Each source (vertex) was normalized by division over the sum of the ASSR power across all 20,484 vertices at a single-subject level, thereby expressing the power of each source as a fraction of the total ASSR power. These values were also averaged across the Low and High voices. The across-subject grand average difference in normalized source power across the cortical space between the nonoverlapping and overlapping experiments (i.e. SD1–SD2) in (a) is shown as a percentage of the normalized overlapping-voices ASSR power in (b) for easier interpretation. All clusters with P < 0.05 obtained from a cluster-based permutation test of the nonoverlapping minus overlapping difference (SD1–SD2) are demarcated in white with labeled corresponding P-values. For visual clarity, the figure in (b) only includes vertices at least 10 times the median SD1–SD2 power. The results suggest that up to 50% more resources were proportionally allocated from the temporal–parietal regions to the frontal regions when 2 simultaneous voices instead of one were processed. Orientation views (clockwise starting from top-left): right lateral, left lateral, frontal, top.

Fig. 6

Effect of bottom-up attention and voice on overlapping-voices ASSR2 power. ANOVA results reported significant main effects of (a) bottom-up attention (P = 0.0011) and (b) voice (P < 0.001), but no significant interaction between them (P = 0.37). The mean lg(power) values are plotted with their equivalents in T² scale displayed in parentheses below. Vertical bars denote 0.95 confidence intervals.

Fig. 7

Interaction between bottom-up (BU) and top-down (TD) attentional modulation of ASSR2 power. A 3-factor ANOVA between TD attention, BU attention and voice revealed significant interaction between TD and BU attention (P = 0.0072). Post-hoc Tukey tests of this interaction revealed that the only significant difference (P = 0.0067) was between the BU attend (left column) and BU unattend (right column), during the TD attend condition. The difference between TD attend and TD unattend when BU attention engaged (BU attend) missed significance when using a 2-tailed test (P = 0.065), although it would have been significant if a 1-tailed test was used instead to test only for the effect of attentional enhancement. The mean lg(power) values are plotted with their equivalents in T² scale displayed in parentheses below. Vertical bars denote 0.95 confidence intervals.