Representations of pitch and slow modulation in auditory cortex

Abstract

Iterated ripple noise (IRN) is a type of pitch-evoking stimulus that is commonly used in neuroimaging studies of pitch processing. When contrasted with a spectrally matched Gaussian noise, it is known to produce a consistent response in a region of auditory cortex that includes an area antero-lateral to the primary auditory fields (lateral Heschl's gyrus). The IRN-related response has often been attributed to pitch, although recent evidence suggests that it is more likely driven by slowly varying spectro-temporal modulations not related to pitch. The present functional magnetic resonance imaging (fMRI) study showed that both pitch-related temporal regularity and slow modulations elicited a significantly greater response than a baseline Gaussian noise in an area that has been pre-defined as pitch-responsive. The region was sensitive to both pitch salience and slow modulation salience. The responses to pitch and spectro-temporal modulations interacted in a saturating manner, suggesting that there may be an overlap in the populations of neurons coding these features. However, the interaction may have been influenced by the fact that the two pitch stimuli used (IRN and unresolved harmonic complexes) differed in terms of pitch salience. Finally, the results support previous findings suggesting that the cortical response to IRN is driven in part by slow modulations, not by pitch.

Keywords: Heschl's gyrus; IRN; iterated ripple noise; pitch; planum temporale; spectro-temporal modulation.

Figure 1

Simulated cochlear representations of IRN (top row) and IRNo (bottom row) in the form of spectrograms. The number of delay and add iterations increases from left to right. The analysis smoothes the representation in both time and frequency domains to remove any fine structure related to pitch. All stimuli were created from the same original sample of Gaussian noise, and the IRNo stimuli on the bottom row are processed versions of the stimuli on the top row (IRN). The color bar shows model output in dB SPL. See Barker et al. (2012) for details of the model.

Figure 2

Schematic representation of the continuous stimulation paradigm used for presentation of stimuli in the MR scanner. The 10-ms ramps of the two sounds in each sequence were overlapped at the 3 dB SPL point (at 5 ms) to produce a stable envelope.

Figure 3

Schematic representation of the subset of stimuli that contribute to the 2 × 2 factorial design. Each cell in the matrix contains two levels of salience except for the “no pitch, no modulation” cell.

Figure 4

Statistical T map from the 2 × 2 factorial ANOVA showing locations of the group-averaged responses for the main effects of slow modulation (blue) and pitch (red), and a conjunction for the two features (pink). The yellow border denotes Te 1.2 (lateral portion of HG) and the black border outlines PT (informed by Westbury et al., 1999). Activation is overlaid onto an average anatomical image made from the 12 individual listeners. The left hemisphere is on the left-hand side of each anatomical image. These images used an uncorrected threshold p < 0.05. This figure demonstrates the patterns of activation across the entire cortex, although the analyses were restricted to a 10-mm sphere centered on the white spots in the middle panel.

Figure 5

Plot of the results of the 2 × 2 factorial design within the pitch-responsive ROI (A) and within medial HG (B). The ordinate measures percentage increase in BOLD activation from baseline. Error bars show standard errors.

Figure 6

Plot of the salience analysis results for IRN and for unres within the spherical pitch ROI. “Activation” refers to the average beta weights: a numerical measure of the effect size. The low salience conditions are represented by the light gray bars and the high salience conditions are represented by the dark gray bars. For the low-salience IRN condition, IRNo₄ has been subtracted from IRN₄ and for the high-salience condition, IRNo₆₄ has been subtracted from IRN₆₄ to remove the effects of slow modulation. Error bars represent 95% confidence intervals.